WO2021174739A1 - 神经网络训练方法及装置、电子设备和存储介质 - Google Patents

神经网络训练方法及装置、电子设备和存储介质 Download PDF

Info

Publication number
WO2021174739A1
WO2021174739A1 PCT/CN2020/100715 CN2020100715W WO2021174739A1 WO 2021174739 A1 WO2021174739 A1 WO 2021174739A1 CN 2020100715 W CN2020100715 W CN 2020100715W WO 2021174739 A1 WO2021174739 A1 WO 2021174739A1
Authority
WO
WIPO (PCT)
Prior art keywords
image block
feature extraction
sample image
target area
block
Prior art date
Application number
PCT/CN2020/100715
Other languages
English (en)
French (fr)
Inventor
王娜
宋涛
刘星龙
黄宁
张少霆
Original Assignee
上海商汤智能科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 上海商汤智能科技有限公司 filed Critical 上海商汤智能科技有限公司
Priority to JP2021574781A priority Critical patent/JP2022537974A/ja
Priority to KR1020217041454A priority patent/KR20220009451A/ko
Publication of WO2021174739A1 publication Critical patent/WO2021174739A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V2201/00—Indexing scheme relating to image or video recognition or understanding
    • G06V2201/03—Recognition of patterns in medical or anatomical images

Definitions

  • This application relates to the field of computer technology, in particular to a neural network training method and device, electronic equipment and storage medium.
  • Machine learning are widely used in the field of image processing. For example, they can be applied to the classification and image detection of ordinary images or three-dimensional images. For example, in the processing of medical images, machine learning methods can be used to determine the type of disease and detect the diseased area.
  • lung medical images for example, lung computed tomography (Computed Tomography, CT)
  • CT computed Tomography
  • GGN ground-glass nodules
  • the embodiments of the present application provide a neural network training method and device, electronic equipment, and storage medium.
  • the embodiment of the application provides a neural network training method.
  • the neural network training method is used to train a neural network model and classify images according to the neural network model obtained by training.
  • the method includes: obtaining the position of the target region in the sample image Information and category information; according to the position information of the target area in the sample image, at least one sample image block is obtained by segmentation; according to the category information, the at least one sample image block is classified to obtain N types of sample image blocks, N It is an integer and N ⁇ 1; the N-type sample image blocks are input into the neural network for training.
  • fine classification of sample image blocks can be obtained, and the neural network can be trained, so that the neural network can perform fine classification of images, and improve classification efficiency and accuracy.
  • the sample image is a medical imaging picture.
  • the acquiring the location information and category information of the target area in the sample image includes: locating the target area on the medical imaging picture to obtain the location information of the target area; A pathology picture associated with a medical image picture, where the pathology picture is a diagnosed picture that includes pathological information; according to the pathological information of each target area on the pathology picture, the target area on the medical image picture is determined Category information.
  • inputting the N types of sample image blocks into the neural network for training includes: inputting any sample image block into the neural network for processing to obtain the category prediction information of the sample image block and Predict the target area; determine the classification loss according to at least the category prediction information and the category information of the sample image block; determine the segmentation loss according to the location information of the prediction target area and the sample image block; according to the classification loss And the segmentation loss, training the neural network.
  • determining the classification loss according to the category prediction information and the category information of the sample image block includes: determining the first category according to the category prediction information and the category information of the sample image block Loss; Determine the second classification loss according to the category prediction information and the category information of the class center of the category to which the sample image block belongs; Perform weighted summation processing on the first classification loss and the second classification loss to obtain The classification loss.
  • the category features of the sample image blocks of the same category can be more concentrated during training, and the feature distance between the category information of the sample image blocks of different categories can be larger, which helps to improve the classification performance and improve the classification accuracy. .
  • determining the segmentation loss according to the position information of the prediction target area and the sample image block includes: according to the number of pixels of the prediction target area in the sample image block. A ratio, determining the first weight of the prediction target area and the second weight of the sample background area in the sample image block; according to the first weight, the second weight, the prediction target area, and the sample image block The location information to determine the segmentation loss.
  • the first weight of the prediction target area and the sample background in the sample image block are determined according to the first proportion of the number of pixels of the prediction target area in the sample image block
  • the second weight of the region includes: determining the second proportion of the sample background area in the sample image block according to the first proportion of the number of pixels of the prediction target region in the sample image block; The second ratio is determined as the first weight, and the first ratio is determined as the second weight.
  • the error of the target area and the error of the non-target area can be balanced, which is conducive to the optimization of network parameters, and improves the training efficiency and training effect.
  • the category information includes: pre-invasive adenocarcinoma atypical adenoma nodules, in situ adenocarcinoma nodules, minimally invasive adenocarcinoma nodules, and invasive adenocarcinoma nodules.
  • the neural network includes a shared feature extraction network, a classification network, and a segmentation network.
  • the method further includes: inputting the image block to be processed into the shared feature extraction network for processing to obtain the to-be-processed image block
  • the target feature of the image block where the shared feature extraction network includes M shared feature extraction blocks, the input feature of the i-th shared feature extraction block includes the output features of the first i-1 shared feature extraction blocks, i and M are Integer and 1 ⁇ i ⁇ M; input the target feature into the classification network for classification processing to obtain the category information of the image block to be processed; input the target feature into the segmentation network for segmentation processing to obtain the The target area in the image block to be processed.
  • the shared feature extraction block of the shared feature extraction network can obtain the output features of all previous shared feature extraction blocks, and input its own output features to all subsequent shared feature extractions. Piece. It can strengthen the gradient flow in the network, alleviate the phenomenon of gradient disappearance, and improve the feature extraction and learning capabilities at the same time, which is conducive to finer classification and segmentation of the input image blocks to be processed. In addition, finer category information and target areas of the image blocks to be processed can be obtained, which improves image processing efficiency.
  • inputting the image block to be processed into the shared feature extraction network for processing to obtain the target feature of the image block to be processed includes: performing a first feature extraction process on the image block to be processed, Obtain the first feature of the image block to be processed; input the first feature into the first shared feature extraction block, obtain the output feature of the first shared feature extraction block, and combine the first shared feature
  • the output features of the extraction block are output to the subsequent M-1 shared feature extraction blocks; the output features of the first j-1 shared feature extraction blocks are input to the j-th shared feature extraction block to obtain the j-th shared feature extraction block
  • the output feature of the block where j is an integer and 1 ⁇ j ⁇ M; the output feature of the M-th shared feature extraction block is subjected to the second feature extraction process to obtain the second feature of the image block to be processed;
  • the second feature is pooled to obtain the target feature.
  • the method further includes: preprocessing the image to be processed to obtain the first image; positioning the target area on the first image to determine the target area in the first image Location information; according to the location information of the target area in the first image, at least one image block to be processed is obtained by segmentation.
  • An embodiment of the application provides a neural network training device, the neural network training device is used to train a neural network model, and classify images according to the neural network model obtained by the training, the device includes: an acquisition module configured to acquire a sample image The location information and category information of the target area in the sample image; the first segmentation module is configured to segment to obtain at least one sample image block according to the location information of the target area in the sample image; The at least one sample image block is classified to obtain N types of sample image blocks, where N is an integer and N ⁇ 1; the training module is configured to input the N types of sample image blocks into the neural network for training.
  • An embodiment of the present application provides an electronic device, including: a processor; a memory configured to store a computer program executable by the processor; wherein the processor is configured to execute the above neural network training method through the computer program.
  • An embodiment of the present application provides a storage medium in which a computer program is stored, and the computer program is configured to execute the above neural network training method when running.
  • An embodiment of the present application provides a computer program, including computer-readable code, and when the computer-readable code runs in an electronic device, the processor execution in the electronic device is configured to implement the neural network training method described above.
  • FIG. 1 is a schematic diagram of a system architecture of a neural network training method provided by an embodiment of the present application
  • Fig. 2 is an implementation flowchart of a neural network training method provided by an embodiment of the present application
  • FIG. 3 is a schematic diagram of an application of a neural network training method provided by an embodiment of the present application
  • FIG. 4 is a schematic diagram of a neural network training device provided by an embodiment of the present application.
  • FIG. 5 is a schematic diagram of an electronic device provided by an embodiment of the present application.
  • Fig. 6 is a schematic diagram of another electronic device provided by an embodiment of the present application.
  • methods such as machine learning are widely used in the field of image processing. For example, they can be applied to the classification and image detection of ordinary images or three-dimensional images.
  • Lung cancer is one of the most common malignant tumors in my country. Its mortality rate is the first among all cancer deaths, whether in urban or rural areas, males or females. Among them, adenocarcinoma accounts for about 40% of all lung cancers. Using medical images (for example, lung CT and low-dose spiral CT) for screening, more and more early lung adenocarcinomas are found and manifested as Ground-Glass Nodule (GGN).
  • GGN Ground-Glass Nodule
  • Adenocarcinoma is divided into Atypical Adenomatous Hyperplasia Of Preinvasive Adenocarcinoma (AAHOPA), Adenocarcinoma In Situ (AIS), Minimally Invasive Adenocarcinoma (MIAcarcinoma) and Invasive Adenocarcinoma (Invasive Adenocarcinoma) , IA).
  • the GGN categories of adenocarcinoma include pre-invasive adenocarcinoma atypical adenocarcinoma nodules, in situ adenocarcinoma nodules, minimally invasive adenocarcinoma nodules and invasive adenocarcinoma nodules.
  • the survival period will decrease significantly, which indicates that early detection and diagnosis is an effective and vital method to reduce patient mortality. Therefore, early detection of aggressive features before surgery will be clinically important and can provide guidance for clinical decision-making.
  • lung medical images for example, lung CT
  • machine learning and other methods can be used to determine the type of disease and detect the diseased area. For example, it is possible to predict whether the image of the input nodule is a malignant tumor or a benign tumor. However, there is no detailed classification of the prediction results in the related technology. .
  • computer-aided diagnosis based on artificial intelligence is a more effective method to assess the invasiveness of nodules, and is expected to play an important role in clinical evaluation tasks.
  • FIG. 1 is a schematic diagram of a system architecture of a neural network training method provided by an embodiment of the present application.
  • the system architecture includes a CT instrument 100, a server 200, a network 300, and a terminal device 400.
  • the CT instrument 100 can be connected to the terminal device 400 through the network 300, and the terminal device 400 is connected to the server 200 through the network 300.
  • the CT instrument 100 can be used to collect CT images, for example, an X-ray CT instrument or a gamma-ray CT instrument, etc.
  • a terminal that can scan a certain thickness of a certain part of the human body.
  • the terminal device 400 may be a device with a screen display function, such as a notebook computer, a tablet computer, a desktop computer, or a dedicated message device.
  • the network 300 may be a wide area network or a local area network, or a combination of the two, and uses wireless links to implement data transmission.
  • the server 200 can cut each pathologically proven lung nodule area in the acquired training medical image into a small image block through the designed three-dimensional classification framework, and then the image The blocks are classified, the training data is obtained, and the training data is input to the neural network for training, so that the neural network finely classifies the training medical image pictures, and the trained neural network model is obtained after the training is completed.
  • the medical image picture may be a CT image of the lungs of a patient or a medical examiner collected by the CT instrument 100 of a hospital, a medical examination center, and the like.
  • the server 200 may obtain the medical image picture collected by the CT machine 100 from the terminal device 400 as a training medical image picture, may also obtain a training medical image picture from the CT machine, and may also obtain a training medical image picture from the Internet.
  • the server 200 may be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server based on cloud technology.
  • Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and network within a wide area network or a local area network to realize the calculation, storage, processing, and sharing of data.
  • the provided artificial intelligence cloud service may include a neural network model, and the neural network is trained based on the finely classified training data, so that the neural network can finely classify medical image pictures.
  • the server 200 receives the medical image picture to be processed (for example, a lung CT image)
  • the medical image picture is classified and segmented according to the trained neural network to obtain a finely classified lesion area. Then, the server 200 returns the obtained finely classified lesion area to the terminal device 400 for display, so that the medical staff can view it.
  • the trained neural network can be sent to the terminal device 400, and the terminal device 400 classifies the collected medical image pictures (eg, lung CT images) , Segmentation, etc., to obtain a finely classified lesion area, and display the obtained finely classified lesion area on its own display screen for medical staff to view.
  • the collected medical image pictures eg, lung CT images
  • Segmentation etc.
  • the system architecture of the neural network training method includes the CT instrument 100, the network 300, and the terminal device 400.
  • the terminal device 400 trains the training medical image pictures to obtain a trained neural network, and then The terminal device 400 performs classification, segmentation and other processing on the collected medical image pictures (such as lung CT images) to obtain a finely classified lesion area, and performs the finely classified lesion area on its own display screen. Display for medical staff to view.
  • the embodiment of the application provides a neural network training method, the method is applied to a neural network training device, the neural network training device may be a server, used to train a neural network model, and classify images according to the trained neural network model .
  • the method provided in the embodiment of the present application may be implemented by a computer program, and when the computer program is executed, each step in the neural network training method provided in the embodiment of the present application is completed.
  • the computer program may be executed by a processor.
  • Fig. 2 is an implementation flowchart of a neural network training method provided by an embodiment of the present application. As shown in Fig. 2, the method includes:
  • Step S11 acquiring location information and category information of the target area in the sample image
  • Step S12 According to the position information of the target area in the sample image, at least one sample image block is obtained by segmentation;
  • Step S13 Classify the at least one sample image block according to the category information to obtain N types of sample image blocks, where N is an integer and N ⁇ 1;
  • Step S14 Input the N-type sample image blocks into the neural network for training.
  • fine classification of sample image blocks can be obtained, and the neural network can be trained, so that the neural network can perform fine classification of images, and improve classification efficiency and accuracy.
  • the neural network training method may be executed by terminal equipment or other processing equipment, where the terminal equipment may be User Equipment (UE), mobile equipment, user terminal, terminal, cellular phone, Cordless phones, personal digital assistants (PDAs), handheld devices, computing devices, in-vehicle devices, wearable devices, etc.
  • UE User Equipment
  • PDA personal digital assistant
  • the neural network training method may be implemented by a processor calling a computer program stored in a memory.
  • the sample image is a medical imaging picture, for example, a lung CT image.
  • the sample image block may be an image block including the target area in the sample image.
  • the sample image may be a three-dimensional medical image that has been annotated (for example, category annotation and segmentation annotation), and the sample image block may be an image block containing nodules in the three-dimensional medical image.
  • the position information and category information of the target area in the sample image can be determined to obtain sample image blocks for training the neural network, and label the sample image blocks.
  • Step S11 may include: locating a target area on a medical imaging picture to obtain position information of the target area; obtaining pathology pictures associated with the medical imaging picture; The pathological information determines the category information of the target area on the medical image picture.
  • the pathology picture is a diagnosed picture that includes pathology information, which can be obtained from a medical image database, or sent to the neural network training device after being manually annotated by a professional such as a doctor on the terminal.
  • the sample image may be resampled to obtain a three-dimensional image with a resolution of 1 ⁇ 1 ⁇ 1. And segment the three-dimensional image.
  • segment the three-dimensional image.
  • the target area for example, the lesion area
  • the target area in the normalized three-dimensional image can be located to obtain the position information of the target area.
  • the location information of the target area may be determined by the convolutional neural network used for positioning, or the location information of the target area may be confirmed by professionals such as doctors, etc.
  • the embodiment of the present application does not limit the positioning method.
  • the medical imaging picture may have related pathological pictures, which can be used to determine the type of the lesion in the medical imaging picture.
  • the type of the lesion may include Ground-Glass Nodule (GGN).
  • GGN Ground-Glass Nodule
  • Adenocarcinoma is divided into Atypical Adenomatous Hyperplasia Of Preinvasive Adenocarcinoma (AAHOPA), Adenocarcinoma In situ (AIS), Minimally Invasive Adenocarcinoma (MIA) and Invasive Adenocarcinoma Cancer (Invasive Adenocarcinoma, IA), the examples of this application do not limit the types of lesions.
  • pathological information of each target area can be obtained based on pathological pictures.
  • pathological pictures may be pictures after professional diagnosis, may have analysis descriptions of each lesion, and may be based on pathological pictures. Obtain the pathological information of each target area, and then determine the category information of each target area on the medical imaging picture.
  • image blocks including the lesion area may be cropped from the medical image picture, that is, sample image blocks are cropped, and N types of sample image blocks are obtained according to the category information of the target area. For example, after statistics on the size of the nodules, the size of the sample image block can be determined as 64 ⁇ 64 ⁇ 64, and after cropping and classification, four types (AAHOPA, AIS, MIA, and IA) sample image blocks can be obtained.
  • the sample image block can be rotated, translated, mirrored, zoomed, etc., and the number of samples can be amplified. Moreover, using the amplified sample image block to train the neural network can improve the generalization ability of the neural network and prevent overfitting. In some embodiments of the present application, the positive and negative samples can also be balanced.
  • pre-invasive adenocarcinoma pre-invasive adenocarcinoma, atypical adenoma hyperplasia, in situ adenocarcinoma, minimally invasive adenocarcinoma, and other benign nodules, and invasive adenocarcinoma, and other malignant nodules
  • a large gap between the number of samples and a smaller number of samples can be amplified by the above method to balance the number of positive and negative samples.
  • the embodiment of the application does not limit the manner of amplifying the number of samples.
  • sample image blocks may be input to the neural network in batches.
  • step S14 may include: inputting any sample image block into the neural network for processing to obtain category prediction information and prediction target area of the sample image block; at least according to the category prediction information and the category of the sample image block Information, determine the classification loss; determine the segmentation loss according to the location information of the prediction target area and the sample image block; and train the neural network according to the classification loss and the segmentation loss.
  • the neural network may include a shared feature extraction network, a classification network, and a segmentation network.
  • the sample image block can be feature extracted through the shared feature extraction network to obtain the sample target feature of the sample image block, and the category prediction information of the sample image block can be obtained through the classification network.
  • the category prediction information may have errors, and the category information and The category labeling information of the sample image block determines the classification loss of the neural network.
  • determining the classification loss according to the category prediction information and the label information of the sample image block includes: determining the first category according to the category prediction information and the label information of the sample image block Loss; Determine the second classification loss according to the category prediction information and the category information of the class center of the category to which the sample image block belongs; Perform weighted summation processing on the first classification loss and the second classification loss to obtain The classification loss.
  • the labeling information of the sample image block may include category labeling information.
  • the category labeling information may be information indicating the category of the nodule in the sample image block.
  • the category prediction information may be category information expressed in the form of a vector, etc. The probability distribution of the image block to be processed represented by the vector belonging to each category can be determined through a probability dictionary, etc., and then the category of the image block to be processed can be determined .
  • the vector of category prediction information may directly represent the probability of the image block to be processed.
  • each element of the vector represents the probability of the category of the image block to be processed.
  • the first classification loss can be determined according to the category prediction information and the category labeling information of the sample image block.
  • the feature distance between the vector of category prediction information and the vector of category labeling information can be determined (for example, Euclidean distance, cosine distance, etc.), and determines a first classification according to the characteristic distance L sm loss, e.g., loss can be calculated according to the first classification softmaxloss loss function L sm.
  • the first classification loss L sm can be determined by the following formula (1):
  • x i represents the category prediction information of the i-th sample image block
  • y i represents the category to which the i-th sample image block belongs
  • n represents the number of categories.
  • m represents the number of sample image blocks input to the neural network in each batch
  • b j represents the offset item of the j-th category.
  • the above-mentioned first classification loss is used for training, which can expand the feature distance between classes of class information of different classes, so that the classification network can distinguish sample image blocks of different classes.
  • the difference between multiple types of nodules in the lung is not obvious (for example, the shape of the nodules of in situ adenocarcinoma and minimally invasive adenocarcinoma is not very different), and the shapes of the two nodules of the same type are different (for example, , Invasive adenocarcinoma and other malignant nodules have different shapes), which results in small feature distances between category information and large intra-class feature distances, resulting in poor classification results of the classification network trained using only the first classification loss L sm .
  • the classification network can be trained through the second classification loss.
  • the category information of the class centers of each category in the multiple sample image blocks can be determined.
  • the category information of the class centers of the multiple sample image blocks can be weighted average, or the category information of the sample image blocks can be aggregated.
  • the embodiment of the present application does not limit the class information of the class center.
  • the second classification loss may be determined according to the category prediction information of the sample image block and the category label information of the category center of the category to which it belongs.
  • the distance between the features may be determined category information category and class prediction information center and determines second classification according to the characteristic distance L ct loss, e.g., loss can be calculated according to a second classification centerloss loss function L ct.
  • Training the classification network through the second classification loss L ct can reduce the intra-class feature distance of the class information of similar sample image blocks, so that the same feature information is more concentrated in the feature space, which is beneficial to determine the class of the sample image block.
  • the second classification loss L ct can be determined by the following formula (2):
  • the first classification loss and the second classification loss can be used to jointly determine the classification loss.
  • the first classification loss and the second classification loss can be weighted and summed to obtain the classification loss.
  • the weight ratio of the first classification loss and the second classification loss is 1:0.8, and the classification loss can be obtained after the weighted summation is performed according to the above weight ratio.
  • the embodiment of the present application does not limit the weight ratio.
  • the category features of the sample image blocks of the same category can be more concentrated in the training, so that the distance between the category information of the sample image blocks of different categories is larger, which helps to improve the classification performance and improve the classification accuracy.
  • the sample target feature can be segmented through a segmentation network to obtain the prediction target region in the sample image block.
  • the prediction target area may have an error
  • the segmentation loss can be determined according to the error between the prediction target area and the labeled target area of the sample image block, and then training is performed through the segmentation loss.
  • determining the segmentation loss according to the annotation information of the prediction target area and the sample image block includes: according to the number of pixels of the prediction target area in the sample image block. A ratio, determining the first weight of the prediction target area and the second weight of the sample background area in the sample image block; according to the first weight, the second weight, the prediction target area, and the sample image block The labeling information of, determines the segmentation loss.
  • the labeling information includes the labelled segmentation area
  • the segmentation loss can be determined directly according to the error between the prediction target area and the labelled segmentation area.
  • the diameter of the nodules is usually between 5 millimeters (mm) and 30mm.
  • the area where the nodules are located and other areas in the sample image block has a large difference in proportion, resulting in pixels between the target area and the non-target area.
  • the imbalance of the number can make the error of the prediction target area account for a small proportion of the segmentation loss, which is not conducive to the optimization and adjustment of the neural network, resulting in low training efficiency and poor training effect.
  • the pixels in the target area and the pixels in the non-target area may be weighted according to the weighting process.
  • the first weight of the prediction target area and the second weight of the sample background area in the sample image block may be determined according to the first proportion of the number of pixels of the prediction target area in the sample image block.
  • the pixels of the above two regions are weighted to balance the loss of the target region and the loss of the non-target region.
  • the first weight of the prediction target area and the sample background in the sample image block are determined according to the first proportion of the number of pixels of the prediction target area in the sample image block
  • the second weight of the region includes: determining the second proportion of the sample background area in the sample image block according to the first proportion of the number of pixels of the prediction target region in the sample image block; The second ratio is determined as the first weight, and the first ratio is determined as the second weight.
  • the sample image block may include a prediction target area and a background area, and the proportion of the number of pixels in the prediction target area can be counted to determine the proportion of the sample background area. For example, if the first ratio of the number of pixels in the prediction target area is 0.2, the second ratio of the number of pixels in the sample background area is 0.8. The embodiment of the application does not limit the first ratio and the second ratio.
  • the second ratio is determined as the first weight of the predicted target area, and the first ratio is determined as the second weight of the sample background area. For example, if the first ratio of the number of pixels in the prediction target area is 0.2, the first weight of the prediction target area is 0.8, and the second ratio of the number of pixels in the sample background area is 0.8, and the second weight of the sample background area is 0.2.
  • the segmentation loss may be determined according to the first weight, the second weight, the prediction target area, and the labeling target area of the sample image block.
  • the segmentation loss can be determined based on the difference between the predicted target area and the target area in the labeling information.
  • the pixels in the predicted target area can be weighted, the weight is the first weight, and the sample background area
  • the pixels are weighted, the weight is the second weight, and the weighted segmentation loss L dc is determined .
  • the segmentation loss L dc can be calculated according to the weightedDiceloss loss function.
  • the segmentation loss L dc can be determined by the following formula (3):
  • the error of the target area and the error of the non-target area can be balanced, which is conducive to the optimization of network parameters, and improves the training efficiency and training effect.
  • the comprehensive network loss of the shared feature extraction network, the segmentation network, and the classification network may be determined according to the classification loss and the segmentation loss.
  • the classification loss and the segmentation loss can be weighted and summed to obtain the comprehensive network loss.
  • the comprehensive network loss L total can be determined according to the following formula (4):
  • L total ⁇ 1 L sm + ⁇ 2 L ct + ⁇ 3 L dc (4);
  • ⁇ 1 represents the weight of L sm
  • ⁇ 2 represents the weight of L ct
  • ⁇ 3 represents the weight of L dc .
  • ⁇ 1 1.2
  • ⁇ 2 0.8
  • ⁇ 3 2.
  • the network parameters of the aforementioned neural network can be adjusted inversely through the integrated network loss.
  • the network parameters can be adjusted through the gradient descent method to optimize the network parameters and improve the accuracy of segmentation and classification.
  • the foregoing training method may be iteratively executed multiple times, and training is performed according to a set learning rate.
  • the learning rate in the first 20 training cycles, the learning rate of 0.001*1.1 x (where x represents the training cycle) can be used for training, and in the subsequent training, it can be used in the 40th, 80th, and 120th training cycles respectively.
  • Medium makes the learning rate halved.
  • the training efficiency can be improved, so that the network parameters can be greatly optimized, and the learning rate can be gradually reduced in the subsequent training, the network parameters can be fine-tuned, the accuracy of the neural network can be improved, and the accuracy of classification processing and segmentation processing can be improved.
  • the training can be completed when the training conditions are met, and the trained shared feature extraction network, segmentation network, and classification network can be obtained.
  • the training condition may include the number of training times, that is, the training condition is satisfied when the preset number of training times is reached.
  • the training condition may include that the integrated network loss is less than or equal to a preset threshold or converges to a preset interval, that is, when the integrated network loss is less than or equal to a preset threshold or converges to a preset interval, the accuracy of the neural network can be considered to satisfy the use Upon request, training can be completed.
  • the embodiment of the application does not limit the training conditions.
  • the trained neural network can be tested after the training is completed.
  • the three-dimensional image block including the nodule area in the lung three-dimensional medical image can be input to the above neural network, and the output segmentation result and the accuracy of the classification result can be counted, for example, compared with the annotation information of the three-dimensional image block to determine the segmentation
  • the accuracy of the result and the classification result can determine the training effect of the neural network. If the accuracy rate is higher than the preset threshold, it can be considered that the training effect is good, and the neural network performance is good, which can be used in the process of obtaining the category of the image block to be processed and segmenting the target area. If the accuracy rate does not reach the preset threshold, the training effect can be considered poor, and other sample image blocks can be used to continue training.
  • the trained neural network can obtain the category and target area of the image block to be processed when the target area and category in the image block to be processed are unknown, or it can be in the category of the image block to be processed. In a known situation, only the target area in the image block to be processed is acquired, or when the target area in the image block to be processed is known, the category of the image block to be processed can also be acquired.
  • the embodiment of the application does not limit the use method of the neural network.
  • the neural network trained by the above training method can be used to determine the lesion area and the lesion category in the image block to be processed.
  • the neural network includes a shared feature extraction network, a classification network, and a segmentation network.
  • the method further includes: inputting the image block to be processed into the shared feature extraction network for processing to obtain the target feature of the image block to be processed, wherein the shared feature
  • the extraction network includes M shared feature extraction blocks.
  • the input feature of the i-th shared feature extraction block includes the output features of the first i-1 shared feature extraction block.
  • i and M are integers and 1 ⁇ i ⁇ M; the target The feature input classification network performs classification processing to obtain the category information of the image block to be processed; the target feature is input to the segmentation network for segmentation processing to obtain the target area in the image block to be processed.
  • the shared feature extraction network is used to obtain the target feature.
  • the shared feature extraction block of the shared feature extraction network can obtain the output features of all previous shared feature extraction blocks, and input its own output features to all subsequent shared feature extraction blocks . It can strengthen the gradient flow in the network, alleviate the phenomenon of gradient disappearance, and improve the feature extraction and learning capabilities at the same time, which is conducive to finer classification and segmentation of the input image blocks to be processed. In addition, finer category information and target areas of the image blocks to be processed can be obtained, which improves image processing efficiency.
  • the image block to be processed may be a partial area in the image to be processed.
  • a partial area can be cropped from the image to be processed, for example, an area including the target object can be cropped.
  • the image to be processed is a medical image picture, and the area including the lesion can be cropped in the medical image picture.
  • the image to be processed may be a three-dimensional medical image of the lung (for example, a CT image of the lung), and the image block to be processed may be a three-dimensional image block of a lesion area (for example, an area with nodules) cut out in the image to be processed.
  • the embodiments of the present application do not impose restrictions on the types of images to be processed and image blocks to be processed.
  • a medical image picture for example, a three-dimensional medical image of the lung
  • the size and resolution of the medical image picture are relatively high, and in the medical image picture, there are many areas of normal tissue, so it can be Preprocess the medical imaging pictures, and process the cropped areas including the lesions to improve the processing efficiency.
  • the method further includes: preprocessing the image to be processed to obtain the first image; positioning the target area on the first image to determine the position information of the target area in the first image ; According to the location information of the target area in the first image, segment to obtain at least one image block to be processed.
  • the image to be processed may be preprocessed first to improve processing efficiency.
  • preprocessing such as resampling and normalization can be performed.
  • the three-dimensional medical image of the lung can be resampled to obtain a three-dimensional image with a resolution of 1 ⁇ 1 ⁇ 1 (that is, each pixel represents the content of a 1mm ⁇ 1mm ⁇ 1mm cube).
  • the size of the resampled three-dimensional image can be cropped.
  • there may be some non-pulmonary areas, and the lung area can be cropped to save calculations and improve processing efficiency.
  • the cropped three-dimensional image can be normalized, and the pixel value of each pixel in the three-dimensional image can be normalized to a value range of 0 to 1, so as to improve processing efficiency.
  • the first image is obtained.
  • the embodiment of the present application does not limit the pretreatment method.
  • the target area in the first image can be detected.
  • the target area in the first image can be detected by a convolutional neural network for position detection.
  • a convolutional neural network may be used to detect a region including nodules in a three-dimensional medical image of the lung.
  • the target area may be cropped to obtain the image block to be processed.
  • the area including nodules in the three-dimensional medical image of the lung may be cropped to obtain the image block to be processed.
  • the size of the image block to be processed can be determined according to the size of the nodule, and cut. Crop to obtain one or more image blocks to be processed.
  • the neural network can be used to determine the category information of the image block to be processed, and segment the target area.
  • the image block to be processed is a three-dimensional medical image of the lung that includes nodules. Image block.
  • the type of nodule in the image block to be processed (for example, AAHOPA, AIS, MIA, and IA) can be determined through a neural network, and the area where the nodule is located can be segmented.
  • the target feature of the image block to be processed can be extracted through a shared feature extraction network for classification and segmentation processing.
  • Inputting the image block to be processed into the shared feature extraction network for processing to obtain the target feature of the image block to be processed may include: performing a first feature extraction process on the image block to be processed to obtain the first feature of the image block to be processed; Input the first shared feature extraction block, obtain the output feature of the first shared feature extraction block, and output the output feature of the first shared feature extraction block to the subsequent M-1 shared feature extraction blocks; the first j- The output feature of a shared feature extraction block is input to the j-th shared feature extraction block to obtain the output feature of the j-th shared feature extraction block; the output feature of the M-th shared feature extraction block is subjected to the second feature extraction process to obtain The second feature of the image block to be processed; pooling is performed on the second feature to obtain the target feature.
  • the first feature extraction process can be performed first, for example, through a network including a three-dimensional convolutional layer (Three Dimensional Convolutional Layer), a batch normalization layer (Normalization) and an activation layer (Activiation Layer)
  • the module performs the first feature extraction process to obtain the first feature.
  • the embodiment of the present application does not limit the network level where the first feature extraction process is performed.
  • the shared feature extraction network may include multiple shared feature extraction blocks, and the shared feature extraction block may include multiple network levels, for example, convolutional layers, activation layers, etc.
  • the embodiments of the present application extract shared features
  • the network level included in the block is not limited.
  • the first feature can be processed through multiple shared feature extraction blocks.
  • the number of shared feature extraction blocks is M
  • the first feature can be input into the first shared feature extraction block, that is, the first shared feature extraction block can use the first feature as the input feature and compare the input feature Perform feature extraction processing to obtain output features.
  • the output feature of the first shared feature extraction block can be shared by all subsequent shared feature extraction blocks, that is, the output feature of the first shared feature extraction block can be shared with subsequent M-1 shared features
  • the extraction block is used as the input feature of the subsequent M-1 shared feature extraction blocks.
  • the input feature of the second shared feature extraction block is the output feature of the first shared feature extraction block.
  • the second shared feature extraction block After the second shared feature extraction block performs feature extraction processing on its input features, it can be The output features are output to the subsequent 3rd to Mth shared feature extraction blocks as the input features of the 3rd to Mth shared feature extraction blocks.
  • the input feature of the third shared feature extraction block is the output feature of the first shared feature extraction block and the output feature of the second shared feature extraction block, and the output of the first shared feature extraction block
  • the features and the output features of the second shared feature extraction block can be input to the third shared feature extraction block (ie, The input feature of the third shared feature extraction block can be the output feature of the first shared feature extraction block and the output feature of the second shared feature extraction block), or the third shared feature extraction block can be directly Both the output feature of the first shared feature extraction block and the output feature of the second shared feature extraction block are used as input features (for example, the third shared feature extraction block may include a feature fusion layer, which can be used for feature fusion processing at this level, Or retain all feature channels, and directly perform subsequent processing on the features of all feature channels, that is, perform subsequent processing on the output features of the first shared feature extraction block and the output features of the second shared feature extraction block), and Perform feature extraction processing on the input features (for example, directly perform feature extraction processing on the features of all feature channels
  • the output features of the first j-1 shared feature extraction block can be used as input features to be input to the first shared feature extraction block.
  • the fused features can be used as the input features of the j-th shared feature extraction block, or directly output the first j-1 shared feature extraction blocks
  • the feature is used as the input feature of the j-th shared feature extraction block (for example, fusion is performed in the j-th shared feature extraction block, or subsequent processing is directly performed on the features of all feature channels, that is, the first j-1 shared features are extracted
  • the output characteristics of the block are processed later).
  • the jth shared feature extraction block can perform feature extraction processing on its input features to obtain the output feature of the jth shared feature extraction block, and use the output feature as the input of the j+1th to the Mth shared feature extraction block feature.
  • the M-th shared feature extraction block may obtain the output feature of the M-th shared feature extraction block according to the output features of the first M-1 shared feature extraction block.
  • the second feature extraction process can be performed through the subsequent network level of the shared feature extraction network.
  • the output of the Nth shared feature extraction block can be performed through a network module including a three-dimensional convolution layer, a batch normalization layer, and an activation layer.
  • the second feature extraction process is performed on the feature to obtain the second feature.
  • the embodiment of the present application does not limit the network level for performing the second feature extraction process.
  • the second feature may be pooled.
  • the second feature may be pooled through an average pooling layer to obtain the target feature.
  • the embodiment of the application does not limit the type of pooling processing.
  • the foregoing processing may be performed multiple times, for example, may include multiple shared feature extraction networks.
  • the first shared feature extraction network can take the first feature as the input feature.
  • the second feature extraction process and the pooling process of the shared feature extraction block the output feature of the first shared feature extraction network is obtained.
  • the two shared feature extraction networks can use the output features of the first shared feature extraction network as input features.
  • the second shared feature extraction can be obtained
  • the output feature of the network can be processed by multiple shared feature extraction networks, and the output feature of the last (for example, the fourth) shared feature extraction network is used as the target feature.
  • the embodiment of the present application does not limit the number of shared feature extraction networks.
  • the shared feature extraction block of the shared feature extraction network can obtain the output features of all previous shared feature extraction blocks, and input its own output features to all subsequent shared feature extractions. Piece. It can strengthen the gradient flow in the network, alleviate the phenomenon of gradient disappearance, and improve the feature extraction and learning capabilities at the same time, which is conducive to finer classification and segmentation of the input image blocks to be processed.
  • the category information of the image block to be processed can be determined according to the target feature.
  • the image block to be processed is an image block including nodules and other lesions in the three-dimensional medical image of the lung, and the nodule can be determined according to the target feature.
  • the category of the section it can be determined whether the type of nodule is pre-invasive adenocarcinoma atypical adenoma hyperplasia, adenocarcinoma in situ, minimally invasive adenocarcinoma, or invasive adenocarcinoma.
  • the target feature can be classified through the classification network to obtain the category information of the image block to be processed.
  • the classification network may include multiple network levels, such as convolutional layer, global average pooling layer (Global Average Pooling), and fully connected layer (Fully Connected Layer), etc.
  • the above network levels can classify target features.
  • And can output category information.
  • the category information may be category information expressed in the form of a vector or the like, and the probability distribution of the image block to be processed represented by the vector belonging to each category can be determined through a probability dictionary or the like, and then the category information of the image block to be processed can be determined.
  • the vector of category information can directly represent the probability of the image block to be processed.
  • each element of the vector represents the probability of the category of the image block to be processed.
  • (0.8, 0.1, 0.1) can represent the image to be processed
  • the probability of a block belonging to the first category is 0.8
  • the probability of belonging to the second category is 0.1
  • the probability of belonging to the third category is 0.1
  • the category with the highest probability can be determined as the category of the image block to be processed, that is, Determine the category information of the image block to be processed as the first category.
  • the embodiment of this application does not limit the representation method of category information.
  • the category information of the image block to be processed can be determined according to the target feature.
  • the image block to be processed is an image block including nodules and other lesions in the three-dimensional medical image of the lung, and the nodule can be determined according to the target feature. The location of the section, and divide the area where it is located.
  • segmentation processing can be performed through a segmentation network to obtain the target area in the image block to be processed, for example, the target area can be segmented.
  • the segmentation network may include multiple network levels, for example, an upsampling layer (Upsample), a fully connected layer, and so on.
  • the target feature is a feature map obtained by performing feature extraction and pooling on the image block to be processed in the shared feature extraction network, and the resolution of the target feature may be lower than that of the image block to be processed.
  • Upsampling can be performed through the upsampling layer to reduce the number of feature channels of the target feature and increase the resolution, so that the feature map output by the segmentation network is consistent with the resolution of the image block to be processed. For example, if the shared feature extraction network performs four pooling processing, four upsampling processing can be performed through the upsampling layer, so that the output feature map of the segmentation network is consistent with the resolution of the image block to be processed.
  • the target area can be segmented in the feature map output by the segmentation network, for example, the target area where the nodule is located is determined by the contour line or the contour surface. The embodiment of the present application does not limit the network hierarchy of the segmented network.
  • the position of the target area in the image to be processed can also be determined.
  • the position of the target area in the image to be processed can be restored according to the position of the image block to be processed in the image to be processed and the position of the target area in the image block to be processed.
  • the position of the nodule in the image block to be processed in the lung medical image, can be segmented, and the position of the nodule in the lung medical image can be restored.
  • fine classification of sample image blocks can be obtained, and the neural network can be trained, so that the neural network can finely classify images, improve classification efficiency and accuracy; and can share features Extract the network to obtain the target feature.
  • the shared feature extraction block of the shared feature extraction network can obtain the output features of all previous shared feature extraction blocks, and input its own output features to all subsequent shared feature extraction blocks to strengthen the gradient flow in the network.
  • FIG. 3 is a schematic diagram of an application of the neural network training method provided by an embodiment of the present application.
  • image blocks As shown in FIG. Nodules) image blocks.
  • the sample image block may have category annotations.
  • the sample image block may include four categories: AAHOPA, AIS, MIA, and IA.
  • the sample image block 32 may be input to the neural network 33.
  • the shared feature extraction network 331 included in the neural network 33 performs feature extraction on each batch of sample image blocks to obtain sample target features of the sample image blocks, and pass
  • the classification network 332 included in the neural network 33 obtains the category prediction information of the sample image block, and the classification loss of the neural network can be determined by formula (1) and formula (2).
  • the segmentation network 333 included in the neural network 33 can obtain the prediction target area in the sample image block 32, and can determine the segmentation loss of the neural network according to formula (3).
  • the weighted sum of the segmentation loss and the classification loss can be used to obtain the comprehensive network loss of the neural network, and the neural network can be trained through the comprehensive network loss.
  • the trained neural network can be used to determine the focus area and focus category in the image block of the medical image.
  • the image to be processed may be a three-dimensional lung medical image (for example, a lung CT image), and the image block to be processed may be a case region (for example, an area with nodules) that is cut out of the image to be processed. ) Of the three-dimensional image block.
  • the three-dimensional medical image can be resampled to obtain a three-dimensional image with a resolution of 1 ⁇ 1 ⁇ 1, and the area where the lungs are located can be cropped, and then the area where the lungs are located can be normalized .
  • the area where the nodule is located in the area where the lung is located can be detected, and a plurality of image blocks to be processed including the area where the nodule is located can be cropped according to a size of 64 ⁇ 64 ⁇ 64.
  • multiple image blocks to be processed may be subjected to feature extraction processing in batches to obtain target features of the image blocks to be processed.
  • the first feature extraction process may be performed first, for example, the first feature extraction process may be performed through a network module including a three-dimensional convolution layer, a batch normalization layer, and an activation layer to obtain the first feature.
  • the first feature may be input to the shared feature extraction network.
  • the shared feature extraction network may include multiple shared feature extraction blocks.
  • the number of shared feature extraction blocks is M
  • the first feature can be input to the first shared feature extraction block for processing
  • the output features of the first shared feature extraction block can be to the subsequent M-1 shared features Extract the block.
  • the input feature of the second shared feature extraction block is the output feature of the first shared feature extraction block
  • the second shared feature extraction block can output its output features to the subsequent 3rd to Mth shared features Extract the block.
  • the input features of the third shared feature extraction block are the output features of the first shared feature extraction block and the output features of the second shared feature extraction block, and the output features of the third shared feature extraction block can be output to the fourth To the M-th shared feature extraction block.
  • the output features of the first j-1 shared feature extraction block can be input to the j-th shared feature extraction block, and the output feature of the j-th shared feature extraction block can be output to the j+1 to M-th shared Feature extraction block.
  • the M-th shared feature extraction block can obtain the output features of the M-th shared feature extraction block according to the output features of the previous M-1 shared feature extraction blocks, and perform the second feature extraction process, for example, by including three-dimensional convolution
  • the network modules of the layer, batch normalization layer, and activation layer perform a second feature extraction process on the output feature of the Nth shared feature extraction block to obtain the second feature.
  • the second feature may be pooled (for example, average pooling) processing to obtain the target feature.
  • the foregoing processing may be performed multiple times (for example, 4 times), for example, multiple shared feature extraction networks may be included.
  • multiple shared feature extraction networks may be included.
  • the classification network may perform classification processing on the target feature to obtain category information of the image block to be processed.
  • the classification network can obtain the category information of the image block to be processed through the convolutional layer, the global average pooling layer, and the fully connected layer.
  • the segmentation network can perform segmentation processing on the target feature to obtain the target area (ie, the area where the nodule is located).
  • the segmentation network performs four upsampling processing through the upsampling layer, so that the output feature map of the segmentation network is consistent with the resolution of the image block to be processed, and the target area can be segmented in the feature map output of the segmentation network .
  • the aforementioned neural network can obtain the category and target area of the image block to be processed when the target area and category in the image block to be processed are unknown (for example, the area where the nodule is located can be segmented, and Obtain the category of the nodule), or when the category of the image block to be processed is known, only obtain the target area in the image block to be processed (for example, segment the area where the nodule is located), or it can be used in the image block to be processed.
  • the category of the image block to be processed is obtained (for example, the category of the nodule is determined).
  • the image processing method can be used to segment and classify case regions in medical images such as lung CT images, improve clinical work efficiency, reduce missed diagnosis and misdiagnosis, and can also be used to classify other images
  • the embodiments of the present application do not limit the application field of the image processing method.
  • FIG. 4 is a schematic diagram of a neural network training device provided by an embodiment of the present application.
  • the device includes: an acquisition module 11 configured to acquire position information and category information of a target area in a sample image; first The segmentation module 12 is configured to segment to obtain at least one sample image block according to the position information of the target area in the sample image; the classification module 13 is configured to classify the at least one sample image block according to the category information to obtain N-type sample image blocks, where N is an integer, and N ⁇ 1; the training module 14 is configured to input the N-type sample image blocks into the neural network for training.
  • the sample image is a medical imaging picture.
  • the acquisition module 11 is further configured to: locate a target area on a medical image picture to obtain location information of the target area; obtain a pathology picture associated with the medical image picture;
  • the pathology picture is a diagnosed picture that includes pathology information; according to the pathology information of each target area on the pathology picture, the category information of the target area on the medical imaging picture is determined.
  • the training module 14 is further configured to: input any sample image block into the neural network for processing to obtain category prediction information and prediction target area of the sample image block; at least according to the category Determine the classification loss according to the prediction information and the category information of the sample image block; determine the segmentation loss according to the location information of the prediction target area and the sample image block; and train the segmentation loss according to the classification loss and the segmentation loss Neural Networks.
  • the training module 14 is further configured to: determine the first classification loss according to the category prediction information and the category information of the sample image block; according to the category prediction information and the sample image Determine the second classification loss based on the category information of the class center of the category to which the block belongs; perform weighted summation processing on the first classification loss and the second classification loss to obtain the classification loss.
  • the training module 14 is further configured to determine the first weight of the prediction target area according to the first proportion of the number of pixels of the prediction target area in the sample image block And the second weight of the sample background area in the sample image block; determine the segmentation loss according to the first weight, the second weight, the prediction target area, and the position information of the sample image block.
  • the training module 14 is further configured to determine the sample background area in the sample image block according to the first proportion of the number of pixels of the prediction target area in the sample image block The second ratio; the second ratio is determined as the first weight, and the first ratio is determined as the second weight.
  • the category information includes: pre-invasive adenocarcinoma atypical adenoma nodules, in situ adenocarcinoma nodules, minimally invasive adenocarcinoma nodules, and invasive adenocarcinoma nodules.
  • the neural network includes a shared feature extraction network, a classification network, and a segmentation network.
  • the device further includes: an obtaining module configured to input the image block to be processed into the shared feature extraction network for processing, Obtain the target feature of the image block to be processed, wherein the shared feature extraction network includes M shared feature extraction blocks, and the input feature of the i-th shared feature extraction block includes the output features of the first i-1 shared feature extraction blocks, i And M are integers and 1 ⁇ i ⁇ M; a classification module configured to input the target feature into the classification network for classification processing to obtain category information of the image block to be processed; a segmentation module configured to divide the target The features are input into the segmentation network to perform segmentation processing to obtain the target region in the image block to be processed.
  • the obtaining module is further configured to: perform a first feature extraction process on the image block to be processed to obtain the first feature of the image block to be processed; and input the first feature into the first feature A shared feature extraction block, the output feature of the first shared feature extraction block is obtained, and the output feature of the first shared feature extraction block is output to the subsequent M-1 shared feature extraction blocks;
  • the output features of j-1 shared feature extraction blocks are input to the j-th shared feature extraction block to obtain the output features of the j-th shared feature extraction block, where j is an integer and 1 ⁇ j ⁇ M;
  • the second feature extraction process is performed on the output features of the two shared feature extraction blocks to obtain the second feature of the image block to be processed; the second feature is pooled to obtain the target feature.
  • the device further includes: a preprocessing module configured to preprocess the image to be processed to obtain a first image; a positioning module configured to locate a target area on the first image, The location information of the target area in the first image is determined; the second segmentation module is configured to segment to obtain at least one image block to be processed according to the location information of the target area in the first image.
  • the functions or modules included in the apparatus provided in the embodiments of the present application may be configured to execute the methods described in the above method embodiments, and for implementation, refer to the description of the above method embodiments.
  • the embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored, and the computer program is configured to execute the above method when running.
  • the computer-readable storage medium may be a non-volatile computer-readable storage medium.
  • An embodiment of the present application further provides an electronic device, including: a processor; a memory configured to store a computer program executable by the processor; wherein the processor is configured to execute the above method through the computer program.
  • the electronic device can be provided as a terminal, server or other form of device.
  • the embodiments of the present application also provide a computer program product, including computer readable code.
  • the processor in the device executes the instructions of the neural network training method provided by any of the above embodiments. .
  • the embodiments of the present application also provide another computer program product configured to store computer-readable instructions, which when executed, cause the computer to perform the operations of the neural network training method provided in any of the foregoing embodiments.
  • Fig. 5 is a schematic diagram of an electronic device provided by an embodiment of the present application.
  • the electronic device 800 may be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and other terminals.
  • the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component 808, an audio component 810, and an input/output (Input/Output, I/O) interface 812 ,
  • the sensor component 814, and the communication component 816 may be a mobile phone, a computer, a digital broadcasting terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, and other terminals.
  • the electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component
  • the processing component 802 generally controls the overall operations of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations.
  • the processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the foregoing method.
  • the processing component 802 may include one or more modules to facilitate the interaction between the processing component 802 and other components.
  • the processing component 802 may include a multimedia module to facilitate the interaction between the multimedia component 808 and the processing component 802.
  • the memory 804 is configured to store various types of data to support operations in the electronic device 800. Examples of these data include instructions for any application or method to operate on the electronic device 800, contact data, phone book data, messages, pictures, videos, etc.
  • the memory 804 can be implemented by any type of volatile or non-volatile storage device or their combination, such as static random access memory (Static Random Access Memory, SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory). Programmable Read Only Memory, EEPROM, Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (Read Only Memory, ROM) , Magnetic memory, flash memory, magnetic disk or optical disk.
  • SRAM static random access memory
  • EPROM Erasable Programmable Read Only Memory
  • PROM Programmable Read Only Memory
  • Read Only Memory Read Only Memory
  • the power supply component 806 provides power for various components of the electronic device 800, and may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.
  • the multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user.
  • the screen may include a liquid crystal display (Liquid Crystal Display, LCD) and a touch panel (TouchPanel, TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user.
  • the touch panel includes one or more touch sensors to sense touch, sliding, and gestures on the touch panel. The touch sensor may not only sense the boundary of a touch or slide action, but also detect the duration and pressure related to the touch or slide operation.
  • the multimedia component 808 includes a front camera and/or a rear camera. When the electronic device 800 is in an operation mode, such as a shooting mode or a video mode, the front camera and/or the rear camera can receive external multimedia data. Each front camera and rear camera can be a fixed optical lens system or have focal length and optical zoom capabilities.
  • the audio component 810 is configured to output and/or input audio signals.
  • the audio component 810 includes a microphone (Microphone, MIC).
  • the microphone is configured to receive an external audio signal.
  • the received audio signal may be stored in the memory 804 or transmitted via the communication component 816.
  • the audio component 810 further includes a speaker configured to output audio signals.
  • the I/O interface 812 provides an interface between the processing component 802 and a peripheral interface module.
  • the above-mentioned peripheral interface module may be a keyboard, a click wheel, a button, and the like. These buttons may include, but are not limited to: home button, volume button, start button, and lock button.
  • the sensor component 814 includes one or more sensors configured to provide the electronic device 800 with various aspects of state evaluation.
  • the sensor component 814 can detect the on/off status of the electronic device 800 and the relative positioning of the components.
  • the component is the display and the keypad of the electronic device 800.
  • the sensor component 814 can also detect the electronic device 800 or the electronic device 800.
  • the position of the component changes, the presence or absence of contact between the user and the electronic device 800, the orientation or acceleration/deceleration of the electronic device 800, and the temperature change of the electronic device 800.
  • the sensor component 814 may include a proximity sensor configured to detect the presence of nearby objects when there is no physical contact.
  • the sensor component 814 may also include a light sensor, such as a complementary metal oxide semiconductor (Complementary Metal Oxide Semiconductor, CMOS) or a charge coupled device (Charge Coupled Device, CCD) image sensor, which can be used in imaging applications.
  • CMOS Complementary Metal Oxide Semiconductor
  • CCD Charge Coupled Device
  • the sensor component 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
  • the communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices.
  • the electronic device 800 can access a wireless network based on a communication standard, such as WiFi, 2G, or 3G, or a combination thereof.
  • the communication component 816 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel.
  • the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communication.
  • NFC Near Field Communication
  • the NFC module can be based on Radio Frequency Identification (RFID) technology, Infrared Data Association (Infrared Data Association, IrDA) technology, Ultra Wide Band (UWB) technology, Bluetooth (Bluetooth, BT) technology and other technologies. Technology to achieve.
  • RFID Radio Frequency Identification
  • IrDA Infrared Data Association
  • UWB Ultra Wide Band
  • Bluetooth Bluetooth, BT
  • the electronic device 800 may be used by one or more application specific integrated circuits (ASIC), digital signal processors (Digital Signal Process, DSP), and digital signal processing equipment (Digital Signal Process).
  • ASIC application specific integrated circuits
  • DSP digital signal processors
  • Digital Signal Process Digital Signal Process
  • DSPD Digital Signal Process
  • PLD Programmable Logic Device
  • FPGA Field Programmable Gate Array
  • controller microcontroller, microprocessor or other electronic components, configured to execute The above method.
  • non-volatile computer-readable storage medium such as the memory 804 including computer program instructions, which can be executed by the processor 820 of the electronic device 800 to complete the foregoing method.
  • Fig. 6 is a schematic diagram of another electronic device provided by an embodiment of the present application.
  • the electronic device 1900 may be provided as a server. 6
  • the electronic device 1900 includes a processing component 1922, and may also include one or more processors, and a memory resource represented by the memory 1932, configured to store instructions executable by the processing component 1922, such as application programs.
  • the application program stored in the memory 1932 may include one or more modules each corresponding to a set of instructions.
  • the processing component 1922 is configured to execute instructions to perform the above-described methods.
  • the electronic device 1900 may also include a power supply component 1926 configured to perform power management of the electronic device 1900, a wired or wireless network interface 1950 configured to connect the electronic device 1900 to the network, and an input output (I/O) interface 1958 .
  • the electronic device 1900 can operate based on an operating system stored in the memory 1932, such as Windows ServerTM, Mac OS XTM, UnixTM, LinuxTM, FreeBSDTM or the like.
  • a non-volatile computer-readable storage medium such as the memory 1932 including computer program instructions, which can be executed by the processing component 1922 of the electronic device 1900 to complete the foregoing method.
  • the computer program product may include a computer-readable storage medium loaded with computer-readable program instructions configured to enable a processor to implement various aspects of the present application.
  • the computer-readable storage medium may be a tangible device that holds and stores instructions used by the instruction execution device.
  • the computer-readable storage medium may be, for example, but not limited to: an electrical storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing.
  • the computer-readable storage medium may include: portable computer disk, hard disk, random access memory (Random Access Memory, RAM), ROM, erasable programmable read-only memory (EPROM or flash memory), SRAM, portable compact disk read-only memory (Compact Disk-Read Only Memory, CD-ROM), Digital Video Disc (DVD), memory sticks, floppy disks, mechanical coding devices, such as punch cards with instructions stored on them or protrusions in the grooves Structure, and any suitable combination of the above.
  • the computer-readable storage medium used here is not interpreted as a transient signal itself, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (such as light pulses through fiber optic cables), or through wires Transmission of electrical signals.
  • the computer-readable program instructions described herein can be downloaded from a computer-readable storage medium to various computing/processing devices, or downloaded to an external computer or external storage device via a network, such as the Internet, a local area network, a wide area network, and/or a wireless network.
  • the network may include copper transmission cables, optical fiber transmission, wireless transmission, routers, firewalls, switches, gateway computers, and/or edge servers.
  • the network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network, and forwards the computer-readable program instructions for storage in the computer-readable storage medium in each computing/processing device .
  • the computer program instructions configured to perform the operations of this application may be assembly instructions, instruction set architecture (Instruction Set Architecture, ISA) instructions, machine instructions, machine-related instructions, microcode, firmware instructions, status setting data, or one or more Source code or object code written in any combination of two programming languages, including object-oriented programming languages, such as Smalltalk, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages.
  • Computer-readable program instructions can be executed entirely on the user's computer, partly on the user's computer, executed as a stand-alone software package, partly on the user's computer and partly executed on a remote computer, or entirely on the remote computer or server implement.
  • the remote computer may be connected to the user's computer through any kind of network (including a local area network or a wide area network), or may be connected to an external computer (for example, using an Internet service provider to connect through the Internet).
  • electronic circuits such as programmable logic circuits, FPGAs, or programmable logic arrays (Programmable Logic Array, PLA), can be customized by using the status information of computer-readable program instructions. Read the program instructions to realize all aspects of this application.
  • These computer-readable program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine that makes these instructions when executed by the processor of the computer or other programmable data processing device , A device that implements the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams is produced. It is also possible to store these computer-readable program instructions in a computer-readable storage medium. These instructions make computers, programmable data processing apparatuses, and/or other devices work in a specific manner. Thus, the computer-readable medium storing the instructions includes An article of manufacture, which includes instructions for implementing various aspects of the functions/actions specified in one or more blocks in the flowcharts and/or block diagrams.
  • each block in the flowchart or block diagram may represent a module, program segment, or part of an instruction, and the module, program segment, or part of an instruction includes one or more components configured to implement the specified logic function.
  • Executable instructions may also occur in a different order from the order marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, or they can sometimes be executed in the reverse order, depending on the functions involved.
  • each block in the block diagram and/or flowchart, and the combination of the blocks in the block diagram and/or flowchart can be implemented by a dedicated hardware-based system that performs the specified functions or actions Or it can be realized by a combination of dedicated hardware and computer instructions.
  • the computer program product can be implemented by hardware, software or a combination thereof.
  • the computer program product may be embodied as a computer storage medium.
  • the computer program product may be embodied as a software product, such as a software development kit (SDK), etc. Wait.
  • SDK software development kit
  • This application relates to a neural network training method and device, electronic equipment, and storage medium.
  • the method includes: obtaining position information and category information of a target area in a sample image; and segmenting to obtain at least one sample image block according to the position information of the target area ; According to the category information, classify at least one sample image block to obtain N-type sample image blocks; input the N-type sample image blocks into the neural network for training.
  • fine classification of sample image blocks can be obtained, and the neural network can be trained, so that the neural network can finely classify images, and improve classification efficiency and accuracy.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Molecular Biology (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Medical Informatics (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)
  • Apparatus For Radiation Diagnosis (AREA)

Abstract

本申请涉及一种神经网络训练方法及装置、电子设备和存储介质，用于训练神经网络模型，根据训练得到的神经网络模型对图像进行分类，所述方法包括：获取样本图像中目标区域的位置信息及类别信息；根据样本图像中目标区域的位置信息，分割得到至少一个样本图像块；根据类别信息，将至少一个样本图像块进行分类，得到N类样本图像块；将N类样本图像块输入至神经网络中进行训练。

Description

神经网络训练方法及装置、电子设备和存储介质
相关申请的交叉引用
本申请基于申请号为202010148544.8、申请日为2020年03月05日的中国专利申请提出，并要求该中国专利申请的优先权，该中国专利申请的全部内容在此引入本申请作为参考。
技术领域
本申请涉及计算机技术领域，尤其涉及一种神经网络训练方法及装置、电子设备和存储介质。
背景技术
机器学习等方法在图像处理领域有着广泛应用，例如，可应用于普通图像或三维图像的分类和图像检测等领域。例如，在医学图像的处理中，可通过机器学习方法来确定患病的类别以及检测病变区域等。
在医学图像的处理中，肺部医学图像(例如，肺部计算机断层扫描(Computed Tomography，CT))的分类和检测对肺炎、肺癌等病变的筛查和诊断起着重要作用。在术前及早发现病变特征在临床上是极为重要的，可为临床决策提供指导。但是，由于缺乏早期癌症的典型放射学特征(气泡清晰，胸膜回缩等)，临床上，专家或放射科医生很难准确地从CT图像上鉴别诊断亚型磨玻璃结节(Ground-Glass Nodule，GGN)类别。
发明内容
本申请实施例提供一种神经网络训练方法及装置、电子设备和存储介质。
本申请实施例提供一种神经网络训练方法，所述神经网络训练方法用于训练神经网络模型，根据训练得到的神经网络模型对图像进行分类，所述方法包括：获取样本图像中目标区域的位置信息及类别信息；根据所述样本图像中目标区域的位置信息，分割得到至少一个样本图像块；根据所述类别信息，将所述至少一个样本图像块进行分类，得到N类样本图像块，N为整数，且N≥1；将所述N类样本图像块输入至神经网络中进行训练。
根据本申请实施例的神经网络训练方法，可获得样本图像块的精细分类，并对神经网络进行训练，使得神经网络可对图像进行精细分类，提高分类效率和准确度。
在本申请一些实施例中，所述样本图像为医学影像图片。
在本申请一些实施例中，所述获取样本图像中目标区域的位置信息及类别信息包括：对所述医学影像图片上的目标区域进行定位，得到所述目标区域的位置信息；获取与所述医学影像图片关联的病理学图片，所述病理学图片为经过诊断的包括病理信息的图片；根据所述病理学图片上的各目标区域的病理信息，确定所述医学影像图片上的目标区域的类别信息。
在本申请一些实施例中，将所述N类样本图像块输入至神经网络中进行训练，包括：将任一的样本图像块输入所述神经网络进行处理，获得样本图像块的类别预测信息和预测目标区域；至少根据所述类别预测信息和所述样本图像块的类别信息，确定分类损失；根据所述预测目标区域和所述样本图像块的位置信息，确定分割损失；根据所述分类损失和所述分割损失，训练所述神经网络。
在本申请一些实施例中，根据所述类别预测信息和所述样本图像块的类别信息，确 定分类损失，包括：根据所述类别预测信息和所述样本图像块的类别信息，确定第一分类损失；根据所述类别预测信息和所述样本图像块所属类别的类中心的类别信息，确定第二分类损失；对所述第一分类损失和所述第二分类损失进行加权求和处理，获得所述分类损失。
通过这种方式，可在训练中使相同类别样本图像块的类别特征更聚集，使不同类别的样本图像块的类别信息之间的特征距离更大，有助于提升分类性能，提高分类准确率。
在本申请一些实施例中，根据所述预测目标区域和所述样本图像块的位置信息，确定分割损失，包括：根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中样本背景区域的第二权重；根据所述第一权重、第二权重、所述预测目标区域和所述样本图像块的位置信息，确定所述分割损失。
在本申请一些实施例中，根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中样本背景区域的第二权重，包括：根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述样本图像块中样本背景区域的第二比例；将所述第二比例确定为所述第一权重，并将所述第一比例确定为第二权重。
通过这种方式，可平衡目标区域的误差和非目标区域的误差，有利于网络参数优化，提升训练效率和训练效果。
在本申请一些实施例中，所述类别信息包括：浸润前腺癌非典型腺瘤增生结节、原位腺癌结节、微创腺癌结节和浸润性腺癌结节。
在本申请一些实施例中，所述神经网络包括共享特征提取网络、分类网络和分割网络，所述方法还包括：将待处理图像块输入所述共享特征提取网络进行处理，获得所述待处理图像块的目标特征，其中，所述共享特征提取网络包括M个共享特征提取块，第i个共享特征提取块的输入特征包括前i-1个共享特征提取块的输出特征，i和M为整数且1＜i≤M；将所述目标特征输入所述分类网络进行分类处理，获得所述待处理图像块的类别信息；将所述目标特征输入所述分割网络进行分割处理，获得所述待处理图像块中的目标区域。
通过这种方式，能够通过共享特征提取网络来获得目标特征，共享特征提取网络的共享特征提取块可获得之前所有共享特征提取块的输出特征，并将自身的输出特征输入至后续所有共享特征提取块。可加强网络内的梯度流动，缓解梯度消失现象，同时提高特征提取和学习能力，有利于对输入的待处理图像块进行更精细地分类和分割处理。并可获得待处理图像块的较精细的类别信息和目标区域，提升图像处理效率。
在本申请一些实施例中，将待处理图像块输入所述共享特征提取网络进行处理，获得所述待处理图像块的目标特征，包括：对所述待处理图像块进行第一特征提取处理，获得所述待处理图像块的第一特征；将所述第一特征输入第一个共享特征提取块，获得所述第一个共享特征提取块的输出特征，并将所述第一个共享特征提取块的输出特征输出至后续的M-1个共享特征提取块；将前j-1个共享特征提取块的输出特征输入至第j个共享特征提取块，获得所述第j个共享特征提取块的输出特征，其中，j为整数且1＜j＜M；将第M个共享特征提取块的输出特征进行第二特征提取处理，获得所述待处理图像块的第二特征；对所述第二特征进行池化处理，获得所述目标特征。
在本申请一些实施例中，所述方法还包括：对待处理图像进行预处理，获得第一图像；对所述第一图像上的目标区域进行定位，确定所述第一图像中的目标区域的位置信息；根据所述第一图像中的目标区域的位置信息，分割得到至少一个所述待处理图像块。
本申请实施例提供一种神经网络训练装置，所述神经网络训练装置用于训练神经网 络模型，根据训练得到的神经网络模型对图像进行分类，所述装置包括：获取模块，配置为获取样本图像中目标区域的位置信息及类别信息；第一分割模块，配置为根据所述样本图像中目标区域的位置信息，分割得到至少一个样本图像块；分类模块，配置为根据所述类别信息，将所述至少一个样本图像块进行分类，得到N类样本图像块，N为整数，且N≥1；训练模块，配置为将所述N类样本图像块输入至神经网络中进行训练。
本申请实施例提供一种电子设备，包括：处理器；配置为存储处理器可执行计算机程序的存储器；其中，所述处理器被配置为：通过所述计算机程序执行上述神经网络训练方法。
本申请实施例提供一种存储介质，所述存储介质中存储有计算机程序，所述计算机程序被配置为运行时执行上述神经网络训练方法。
本申请实施例提供一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，电子设备中的处理器执行被配置为实现上述神经网络训练方法。
应当理解的是，以上的一般描述和后文的细节描述仅是示例性和解释性的，而非限制本申请。根据下面参考附图对示例性实施例的详细说明，本申请的其它特征及方面将变得清楚。
附图说明
此处的附图被并入说明书中并构成本说明书的一部分，这些附图示出了符合本申请的实施例，并与说明书一起用于说明本申请的技术方案。
图1是本申请实施例提供的神经网络训练方法的系统架构示意图；
图2是本申请实施例提供的神经网络训练方法的一种实现流程图；
图3是本申请实施例提供的神经网络训练方法的一种应用示意图；
图4是本申请实施例提供的神经网络训练装置的一种示意图；
图5是本申请实施例提供的一种电子设备的示意图；
图6是本申请实施例提供的另一种电子设备的示意图。
具体实施方式
以下将参考附图详细说明本申请的各种示例性实施例、特征和方面。附图中相同的附图标记表示功能相同或相似的元件。尽管在附图中示出了实施例的各种方面，但是除非特别指出，不必按比例绘制附图。
在这里专用的词“示例性”意为“用作例子、实施例或说明性”。这里作为“示例性”所说明的任何实施例不必解释为优于或好于其它实施例。
本文中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。另外，本文中术语“至少一种”表示多种中的任意一种或多种中的至少两种的任意组合，例如，包括A、B、C中的至少一种，可以表示包括从A、B和C构成的集合中选择的任意一个或多个元素。
另外，为了更好的说明本申请，在下文的实施方式中给出了众多的细节。本领域技术人员应当理解，没有某些细节，本申请同样可以实施。在一些实例中，对于本领域技术人员熟知的方法、手段、元件和电路未作详细描述，以便于凸显本申请的主旨。
对本申请实施例进行进一步详细说明之前，先对相关技术中的缺陷进行说明。
在相关技术中，机器学习等方法在图像处理领域有着广泛应用，例如，可应用于普通图像或三维图像的分类和图像检测等领域。
肺癌是我国最常见的恶性肿瘤之一，其死亡率无论是在城市或农村、男性或女性， 均居癌症死亡的首位，其中，腺癌约占所有肺癌的40％。使用医学图像(例如，肺部CT和低剂量螺旋CT)进行筛查，越来越多的早期肺腺癌被发现并表现为磨玻璃结节(Ground-Glass Nodule，GGN)，腺癌分为浸润前腺癌非典型腺瘤增生(Atypical Adenomatous Hyperplasia Of Preinvasive Adenocarcinoma，AAHOPA)，原位腺癌(Adenocarcinoma In Situ，AIS)，微创腺癌(Minimally Invasive Adenocarcinoma，MIA)和浸润性腺癌(Invasive Adenocarcinoma，IA)。腺癌的GGN类别包括浸润前腺癌非典型腺瘤增生结节、原位腺癌结节、微创腺癌结节和浸润性腺癌结节。随着肿瘤大小的增加，生存期会出现显著下降，这表明早期发现和诊断是降低患者死亡率的有效且至关重要的方法。因此，在手术前及早发现侵袭性特征在临床上将是重要的，并可为临床决策提供指导。
在医学图像处理中，肺部医学图像(例如，肺部CT)的分类和检测在医学筛查和肺炎、肺癌等诊断的诊断中具有重要作用。在相关技术中，可以通过机器学习等方式来确定患病的类别以及检测病变区域等，例如预测输入的结节的图像属于恶性肿瘤还是良性肿瘤，但是，相关技术中没有对预测结果做细分类。并且，由于缺乏早期癌症的典型放射学特征(气泡清晰，胸膜回缩等)，临床上，专家或放射科医生很难准确地从CT图像上鉴别诊断亚型GGN类别。在这种情况下，基于人工智能的计算机辅助诊断是评估结节侵袭性的一种更加有效方法，有望在临床评估任务中发挥重要作用。
参见图1，图1是本申请实施例提供的神经网络训练方法的系统架构示意图，如图1所示，该系统架构中包括，CT仪100、服务器200、网络300和终端设备400，为实现支撑一个示例性应用，CT仪100可通过网络300连接终端设备400，终端设备400通过网络300连接服务器200，CT仪100可用于采集CT图像，例如可以是X射线CT仪或γ射线CT仪等可对人体某部一定厚度的层面进行扫描的终端。终端设备400可以是笔记本电脑，平板电脑，台式计算机，专用消息设备等具有屏幕显示功能的设备。网络300可以是广域网或者局域网，又或者是二者的组合，使用无线链路实现数据传输。
服务器200可以基于本申请实施例提供的神经网络训练方法，通过设计的三维分类框架，将获取的训练医学影像图片中每个经过病理证明的肺结节区域剪裁为小的图像块，再对图像块进行分类，得到训练数据，将训练数据输入至神经网络进行训练，使得神经网络对训练医学影像图片进行精细分类，训练完成后得到训练好的神经网络模型。医学影像图片可以是医院、体检中心等机构的CT仪100采集的病人或体检人员的肺部CT图像。服务器200可以从终端设备400获取由CT仪100采集的医学影像图片作为训练医学影像图片，也可以从CT仪获取训练医学影像图片，还可以从网络上获取训练医学影像图片。
服务器200可以是独立的物理服务器，也可以是多个物理服务器构成的服务器集群或者分布式系统，还可以是基于云技术的云服务器。云技术是指在广域网或局域网内将硬件、软件、网络等系列资源统一起来，实现数据的计算、存储、处理和共享的一种托管技术。在本申请实施例中，当服务器200为云服务器时，提供的人工智能云服务可以包括神经网络模型，并基于精细分类的训练数据训练神经网络，以使神经网络对医学影像图片进行精细分类。
作为示例，服务器200在接收到待处理的医学影像图片(如，肺部CT图像)后，根据训练好的神经网络对医学影像图片进行分类、分割等处理，得到精细分类的病灶区域。然后，服务器200将得到的精细分类的病灶区域返回给终端设备400进行显示，以便医护人员查看。
在本申请一些实施例中，服务器200训练完成后，可以将训练好的神经网络发送至终端设备400，由终端设备400对采集的待处理的医学影像图片(如，肺部CT图像) 进行分类、分割等处理，得到精细分类的病灶区域，并将得到的精细分类的病灶区域在自身的显示屏上进行显示，以便医护人员查看。
在本申请的一些实施例中，神经网络训练方法的系统架构中包括CT仪100、网络300和终端设备400，由终端设备400对训练医学影像图片进行训练，得到训练好的神经网络，再由终端设备400对采集的待处理的医学影像图片(如，肺部CT图像)进行分类、分割等处理，得到精细分类的病灶区域，并将得到的精细分类的病灶区域在自身的显示屏上进行显示，以便医护人员查看。
本申请实施例提供一种神经网络训练方法，所述方法应用于神经网络训练装置，所述神经网络训练装置可以是服务器，用于训练神经网络模型，根据训练得到的神经网络模型对图像进行分类。本申请实施例提供的方法可以通过计算机程序来实现，该计算机程序在执行的时候，完成本申请实施例提供的神经网络训练方法中各个步骤。在一些实施例中，该计算机程序可以被处理器执行。图2是本申请实施例提供的神经网络训练方法的一种实现流程图，如图2所示，所述方法包括：
步骤S11，获取样本图像中目标区域的位置信息及类别信息；
步骤S12，根据所述样本图像中目标区域的位置信息，分割得到至少一个样本图像块；
步骤S13，根据所述类别信息，将所述至少一个样本图像块进行分类，得到N类样本图像块，N为整数，且N≥1；
步骤S14，将所述N类样本图像块输入至神经网络中进行训练。
根据本申请实施例提供的神经网络训练方法，可获得样本图像块的精细分类，并对神经网络进行训练，使得神经网络可对图像进行精细分类，提高分类效率和准确度。
在本申请一些实施例中，所述神经网络训练方法可以由终端设备或其它处理设备执行，其中，终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、蜂窝电话、无绳电话、个人数字处理(Personal Digital Assistant，PDA)、手持设备、计算设备、车载设备、可穿戴设备等。其它处理设备可为服务器或云端服务器等。在本申请一些实施例中，该神经网络训练方法可以通过处理器调用存储器中存储的计算机程序的方式来实现。
在本申请一些实施例中，所述样本图像为医学影像图片，例如，肺部CT图像等。样本图像块可以是样本图像中包括目标区域的图像块。在示例中，样本图像可以是经过标注(例如，类别标注和分割标注)的三维医学图像，样本图像块可以是三维医学图像中包含结节的图像块。
在本申请一些实施例中，在步骤S11中，可确定样本图像中目标区域的位置信息和类别信息，以获取用于训练神经网络的样本图像块，并对样本图像块进行标注。步骤S11可包括：对医学影像图片上的目标区域进行定位，得到所述目标区域的位置信息；获取与所述医学影像图片关联的病理学图片；根据所述病理学图片上的各目标区域的病理信息，确定所述医学影像图片上的目标区域的类别信息。所述病理学图片为经过诊断的包括病理信息的图片，可以从医学图像数据库获取，或者由医生等专业人员在终端手动标注后发送给神经网络训练装置。
在本申请一些实施例中，可对样本图像进行重采样处理，获得分辨率为1×1×1的三维图像。并对该三维图像进行分割，例如，在肺部三维医学图像中，可能存在部分肺实质以外区域，而肺结节等病灶往往存在肺实质内，故剪裁出(即分割)肺实质所在图像块，并将该图像块进行归一化处理。并可对归一化处理后的三维图像中的目标区域(例如，病灶区域)进行定位，得到目标区域的位置信息。例如，可通过用于定位的卷积神经网络确定目标区域的位置信息，或通过医生等专业人员确认目标区域的位置信息等， 本申请实施例对定位方式不做限制。
在本申请一些实施例中，医学影像图片可具有相关的病理学图片，可用于确定医学影像图片中病灶的类别，例如，病灶的类别可包括磨玻璃结节(Ground-Glass Nodule，GGN)。腺癌分为浸润前腺癌非典型腺瘤增生(Atypical Adenomatous Hyperplasia Of Preinvasive Adenocarcinoma，AAHOPA)，原位腺癌(Adenocarcinoma In Situ，AIS)，微创腺癌(Minimally Invasive Adenocarcinoma，MIA)和浸润性腺癌(Invasive Adenocarcinoma，IA)，本申请实施例对病灶的类别不做限制。
在本申请一些实施例中，可根据病理学图片，获得各目标区域的病理信息，例如，病理学图片可以是经过专业诊断后的图片，可具有对各病灶的分析描述，可根据病理学图片获得各目标区域的病理信息，进而确定医学影像图片上各目标区域的类别信息。
在本申请一些实施例中，可在医学影像图片中剪裁出包括病灶区域的图像块，即，剪裁出样本图像块，并根据目标区域的类别信息，获得N类样本图像块。例如，经过对结节尺寸的统计，可将样本图像块的尺寸确定为64×64×64，经过剪裁和分类，获得四类(AAHOPA、AIS、MIA和IA)样本图像块。
在本申请一些实施例中，由于医学影像图片数量较少，且标注难度大，成本高，而如果将三维图像拆分为多个二维图像，则会损失空间信息，导致性能下降。可将样本图像块进行旋转、平移、镜像、缩放等操作，可扩增样本数量，并且，使用扩增的样本图像块训练神经网络，可提升神经网络的泛化能力，防止过拟合。在本申请一些实施例中，还可平衡正负样本，在示例中，浸润前腺癌非典型腺瘤增生、原位腺癌、微创腺癌等良性结节和浸润性腺癌等恶性结节的样本数量有较大差距，可通过上述方法扩增数量较少的样本，使得正负样本数量平衡。本申请实施例对扩增样本数量的方式不做限制。
在本申请一些实施例中，可分批将样本图像块输入神经网络。其中，步骤S14可包括：将任一的样本图像块输入所述神经网络进行处理，获得样本图像块的类别预测信息和预测目标区域；至少根据所述类别预测信息和所述样本图像块的类别信息，确定分类损失；根据所述预测目标区域和所述样本图像块的位置信息，确定分割损失；根据所述分类损失和所述分割损失，训练所述神经网络。
在本申请一些实施例中，所述神经网络可包括共享特征提取网络、分类网络和分割网络。可通过共享特征提取网络对样本图像块进行特征提取，获得样本图像块的样本目标特征，并通过分类网络获得样本图像块的类别预测信息，类别预测信息可能存在误差，可通过样本预测类别信息和样本图像块的类别标注信息，确定神经网络的分类损失。
在本申请一些实施例中，根据所述类别预测信息和所述样本图像块的标注信息，确定分类损失，包括：根据所述类别预测信息和所述样本图像块的标注信息，确定第一分类损失；根据所述类别预测信息和所述样本图像块所属类别的类中心的类别信息，确定第二分类损失；对所述第一分类损失和所述第二分类损失进行加权求和处理，获得所述分类损失。
在本申请一些实施例中，样本图像块的标注信息可包括类别标注信息，例如，类别标注信息可以是表示样本图像块中的结节的类别的信息。在示例中，所述类别预测信息可以是通过向量等形式表示的类别信息，可通过概率词典等确定该向量表示的待处理图像块属于各类别的概率分布，进而确定待处理图像块的所属类别。或者，类别预测信息的向量可直接表示待处理图像块的概率，在示例中，该向量的各元素分别表示待处理图像块所属类别的概率。
在本申请一些实施例中，可根据类别预测信息和样本图像块的类别标注信息，确定第一分类损失，例如，可确定类别预测信息的向量与类别标注信息的向量之间特征距离(例如，欧氏距离、余弦距离等)，并根据特征距离来确定第一分类损失L sm，例如，可 根据softmaxloss损失函数来计算第一分类损失L sm。在示例中，可通过以下公式(1)来确定第一分类损失L sm：
Figure PCTCN2020100715-appb-000001
其中，x i表示第i个样本图像块的类别预测信息，y i表示第i个样本图像块所属的类别，n表示类别数量。
Figure PCTCN2020100715-appb-000002
表示全连接层中第y i个类别的权重，
Figure PCTCN2020100715-appb-000003
表示全连接层中第j个类别的权重，m表示每个批次输入神经网络的样本图像块的数量，
Figure PCTCN2020100715-appb-000004
表示第i个样本图像块所属的类别偏置项，b j表示第j个类别的偏置项。
在本申请一些实施例中，使用上述第一分类损失进行训练，可扩大不同类别的类别信息的类间特征距离，进而使分类网络可区分不同类别的样本图像块。然而，肺部的多个类别结节之间的差异不明显(例如，原位腺癌和微创腺癌的结节的形状差异并不大)，二同类结节之间形状各异(例如，浸润性腺癌等恶性结节的形状各异)，因此造成了类别信息类间特征距离小，类内特征距离大，导致只使用第一分类损失L sm训练后的分类网络的分类效果不佳。
在本申请一些实施例中，针对上述问题，可通过第二分类损失训练分类网络。在示例中，可确定多个样本图像块中各类别的类中心的类别信息，例如，可对多个样本图像块的类中心的类别信息进行加权平均，或者对样本图像块的类别信息进行聚类处理，获得类中心特征等，本申请实施例对类中心的类别信息不做限制。
在本申请一些实施例中，可根据样本图像块的类别预测信息和其所属类别的类中心的类别标注信息，确定第二分类损失。例如，可确定类别预测信息和类中心的类别信息之间的特征距离并根据特征距离来确定第二分类损失L ct，例如，可根据centerloss损失函数来计算第二分类损失L ct。通过第二分类损失L ct训练分类网络，可缩小同类样本图像块的类别信息的类内特征距离，使得同类的特征信息在特征空间中更集中，有利于确定样本图像块的类别。在示例中，可通过以下公式(2)来确定第二分类损失L ct：
Figure PCTCN2020100715-appb-000005
其中，
Figure PCTCN2020100715-appb-000006
为第i个样本图像块所属的类别的类中心的类别标注信息。
在本申请一些实施例中，可通多第一分类损失和第二分类损失来共同确定分类损失。例如，可将第一分类损失和第二分类损失进行加权求和处理，获得分类损失。例如，第一分类损失和第二分类损失的权重比为1:0.8，按照上述权重比进行加权求和后，可获得分类损失。本申请实施例对权重比不做限制。
通过这种方式，可在训练中使相同类别的样本图像块的类别特征更聚集，使不同类别的样本图像块的类别信息之间的距离更大，有助于提升分类性能，提高分类准确率。
在本申请一些实施例中，可通过分割网络对样本目标特征进行分割处理，获得样本图像块中的预测目标区域。该预测目标区域可具有误差，可根据预测目标区域和所述样本图像块的标注目标区域之间的误差确定分割损失，进而通过分割损失进行训练。
在本申请一些实施例中，根据所述预测目标区域和所述样本图像块的标注信息，确定分割损失，包括：根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中样本背景区域的第二权重；根据所述第一权重、第二权重、所述预测目标区域和所述样本图像块的标注信息，确定所述分割损失。
在本申请一些实施例中，所述标注信息包括标注的分割区域，可直接根据预测目标区域与标注的分割区域之间的误差来确定分割损失。但通常结节的直径为5毫米(millimeter，mm)至30mm之间，样本图像块中结节所在区域和其他区域之间所占比例差距较大，导致目标区域和非目标区域之间的像素数量不平衡，可使得预测目标区域的误差在分割损失中所占比例较小，不利于神经网络的优化调节，使得训练效率较低，训练效果较差。
在本申请一些实施例中，可根据对目标区域的像素和非目标区域(即，样本背景区域)的像素进行加权处理。在示例中，可根据预测目标区域的像素数量在样本图像块中所占的第一比例，确定预测目标区域的第一权重和样本图像块中样本背景区域的第二权重。进而在确定分割损失时，对上述两种区域的像素进行加权处理，来平衡目标区域的损失和非目标区域的损失。
在本申请一些实施例中，根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中样本背景区域的第二权重，包括：根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述样本图像块中样本背景区域的第二比例；将所述第二比例确定为所述第一权重，并将所述第一比例确定为第二权重。
在本申请一些实施例中，样本图像块中可包括预测目标区域和背景区域，可统计预测目标区域的像素数量所占比例，进而确定样本背景区域所占比例。例如，预测目标区域像素数量所占第一比例为0.2，则样本背景区域像素数量所占第二比例为0.8。本申请实施例对第一比例和第二比例不做限制。
在本申请一些实施例中，为使得预测目标区域和样本背景区域平衡，将第二比例确定为预测目标区域的第一权重，并将第一比例确定为样本背景区域的第二权重。例如，预测目标区域像素数量所占第一比例为0.2，则预测目标区域的第一权重为0.8，样本背景区域像素数量所占第二比例为0.8，则样本背景区域的第二权重为0.2。
在本申请一些实施例中，可根据第一权重、第二权重、预测目标区域和样本图像块的标注目标区域，确定分割损失。在示例中，可根据预测目标区域和标注信息中的目标区域的差异，确定分割损失，例如，可将预测目标区域中的像素点进行加权，权重为第一权重，并将样本背景区域中的像素点进行加权，权重为第二权重，并确定加权后的分割损失L dc。例如，可根据weightedDiceloss损失函数来计算分割损失L dc。在示例中，可通过以下公式(3)确定分割损失L dc：
Figure PCTCN2020100715-appb-000007
其中，y k∈{0,1}，y k＝1时，表示第k个像素位置为预测目标区域，y k＝0时，表示第k个像素位置为样本背景区域，P(y k＝1|W)表示分割网络在第k个像素位置的输出，W表示第一权重，Y表示第k个像素位置的分割标注。
通过这种方式，可平衡目标区域的误差和非目标区域的误差，有利于网络参数优化，提升训练效率和训练效果。
在本申请一些实施例中，可根据分类损失和分割损失确定共享特征提取网络、分割网络和分类网络的综合网络损失。例如，可将分类损失和分割损失进行加权求和处理，获得综合网络损失，在示例中，可根据以下公式(4)确定综合网络损失L total：
L total＝θ 1L sm+θ 2L ct+θ 3L dc       (4)；
其中，θ 1表示L sm的权重，θ 2表示L ct的权重，θ 3表示L dc的权重，例如，θ 1＝1.2，θ 2＝0.8， θ 3＝2，本申请实施例对分类损失和分割损失的权重不做限制。
在本申请一些实施例中，可通过综合网络损失反向调节上述神经网络的网络参数，例如，可通过梯度下降法来调节网络参数，使得网络参数优化，提升分割和分类准确率。
在本申请一些实施例中，上述训练方法可迭代执行多次，并根据设定的学习率进行训练。在示例中，在前20个训练周期，可使用0.001*1.1 x(其中x表示训练周期)的学习率进行训练，在随后的训练中，可分别在第40、80和120个……训练周期中使得学习率减半。可在训练的初始阶段，提高训练效率，使得网络参数大幅优化，并在后续的训练中逐步降低学习率，精细调节网络参数，提高神经网络的精度，提高分类处理和分割处理的准确率。
在本申请一些实施例中，可在满足训练条件时，完成训练，获得训练后的共享特征提取网络、分割网络和分类网络。所述训练条件可包括训练次数，即，在达到预设训练次数时，满足训练条件。所述训练条件可包括综合网络损失小于或等于预设阈值或收敛于预设区间，即，当综合网络损失小于或等于预设阈值或收敛于预设区间时，可认为神经网络的精度满足使用要求，可完成训练。本申请实施例对训练条件不做限制。
在本申请一些实施例中，可在训练完成后，对训练后的神经网络进行测试。例如，可将肺部三维医学图像中包括结节区域的三维图像块输入上述神经网络，并统计输出的分割结果和分类结果的准确率，例如，与三维图像块的标注信息进行比较，确定分割结果和分类结果的准确率，即可确定神经网络的训练效果。如果准确率高于预设阈值，可认为训练效果较好，神经网络性能较好，可用于获取待处理图像块的类别并分割出目标区域的处理中。如果准确率未达到预设阈值，可认为训练效果不佳，可使用其他样本图像块继续训练。
在本申请一些实施例中，训练后的神经网络可在待处理图像块中目标区域和类别均未知的情况下，获得待处理图像块的类别和目标区域，也可在待处理图像块的类别已知的情况下，仅获取待处理图像块中的目标区域，或者还可在待处理图像块中目标区域已知的情况下，获取待处理图像块的类别。本申请实施例对神经网络的使用方法不做限制。
在本申请一些实施例中，通过上述训练方法训练的神经网络可用于确定待处理图像块中的病灶区域和病灶类别。所述神经网络包括共享特征提取网络、分类网络和分割网络，所述方法还包括：将待处理图像块输入共享特征提取网络进行处理，获得待处理图像块的目标特征，其中，所述共享特征提取网络包括M个共享特征提取块，第i个共享特征提取块的输入特征包括前i-1个共享特征提取块的输出特征，i和M为整数且1＜i≤M；将所述目标特征输入分类网络进行分类处理，获得所述待处理图像块的类别信息；将所述目标特征输入分割网络进行分割处理，获得所述待处理图像块中的目标区域。
通过这种方式，采用共享特征提取网络来获得目标特征，共享特征提取网络的共享特征提取块可获得之前所有共享特征提取块的输出特征，并将自身的输出特征输入至后续所有共享特征提取块。可加强网络内的梯度流动，缓解梯度消失现象，同时提高特征提取和学习能力，有利于对输入的待处理图像块进行更精细分类和分割处理。并可获得待处理图像块的较精细的类别信息和目标区域，提升图像处理效率。
在本申请一些实施例中，所述待处理图像块可以是待处理图像中的部分区域。在示例中，可从待处理图像中剪裁出部分区域，例如，剪裁出包括目标对象的区域。例如，待处理图像是医学影像图片，可在医学影像图片中剪裁出包括病灶的区域。例如，待处理图像可以是肺部三维医学图像(例如，肺部CT图像)，待处理图像块可以是待处理图像中剪裁出的病灶区域(例如，具有结节的区域)的三维图像块。本申请实施例对待处理图像和待处理图像块的类型不做限制。
在本申请一些实施例中，在医学影像图片(例如，肺部三维医学图像)中，医学影 像图片的尺寸和分辨率较高，且在医学影像图片中，正常组织的区域较多，因此可对医学影像图片进行预处理，并对剪裁出包括病灶的区域进行处理，以提高处理效率。
在本申请一些实施例中，所述方法还包括：对待处理图像进行预处理，获得第一图像；对第一图像上的目标区域进行定位，确定所述第一图像中的目标区域的位置信息；根据所述第一图像中的目标区域的位置信息，分割得到至少一个所述待处理图像块。
在本申请一些实施例中，可首先对待处理图像进行预处理，以提升处理效率。例如，可进行重采样、归一化等预处理。在示例中，可对肺部三维医学图像进行重采样处理，获得分辨率为1×1×1(即，每个像素表示1mm×1mm×1mm的立方体的内容)的三维图像。并可将重采样后的三维图像的尺寸进行剪裁，例如，在肺部三维医学图像中，可能存在部分非肺部区域，可剪裁出肺部所在区域，以节省计算量，提高处理效率。
在示例中，可将剪裁后的三维图像进行归一化，将三维图像中各像素的像素值归一化到0到1的值域范围内，以提升处理效率。在进行归一化处理后，获得所述第一图像。本申请实施例对预处理的方法不做限制。
在本申请一些实施例中，可检测第一图像中的目标区域。例如，可通过用于位置检测的卷积神经网络来检测第一图像中的目标区域。在示例中，可利用卷积神经网络检测肺部三维医学图像中的包括结节的区域。
在本申请一些实施例中，可对目标区域进行剪裁，获得待处理图像块，例如，可对肺部三维医学图像中的包括结节的区域进行剪裁，获得待处理图像块。在示例中，可根据结节的尺寸来确定待处理图像块的尺寸，并进行剪裁，例如，经过对结节尺寸的统计，可将待处理图像块的尺寸确定为64×64×64，经过剪裁，获得一个或多个待处理图像块。
在本申请一些实施例中，可通过所述神经网络来确定待处理图像块的类别信息，并分割出目标区域，例如，待处理图像块为肺部三维医学图像中剪裁出的包括结节的图像块。可通过神经网络确定待处理图像块中结节的种类(例如，AAHOPA、AIS、MIA和IA)，并分割出结节所在的区域。
在本申请一些实施例中，可通过共享特征提取网络来提取待处理图像块的目标特征，以用于分类和分割处理。将待处理图像块输入共享特征提取网络进行处理，获得待处理图像块的目标特征，可包括：对待处理图像块进行第一特征提取处理，获得待处理图像块的第一特征；将第一特征输入第一个共享特征提取块，获得第一个共享特征提取块的输出特征，并将第一个共享特征提取块的输出特征输出至后续的M-1个共享特征提取块；将前j-1个共享特征提取块的输出特征输入至第j个共享特征提取块，获得第j个共享特征提取块的输出特征；将第M个共享特征提取块的输出特征进行第二特征提取处理，获得待处理图像块的第二特征；对第二特征进行池化处理，获得所述目标特征。
在本申请一些实施例中，可首先进行第一特征提取处理，例如，可通过包括三维卷积层(Three Dimensional Convolutional Layer)、批归一化层(Normalization)和激活层(Activiation Layer)的网络模块来进行第一特征提取处理，获得第一特征。本申请实施例对进行第一特征提取处理的网络层级不做限制。
在本申请一些实施例中，共享特征提取网络可包括多个共享特征提取块，共享特征提取块可包括多个网络层级，例如，卷积层、激活层等，本申请实施例对共享特征提取块包括的网络层级不做限制。可通过多个共享特征提取块对第一特征进行处理。在示例中，共享特征提取块的数量为M个，可将第一特征输入第一个共享特征提取块，即，第一个共享特征提取块可将第一特征作为输入特征，并对输入特征进行特征提取处理，获得输出特征，第一个共享特征提取块的输出特征可由后续所有共享特征提取块共享，即，第一个共享特征提取块的输出特征可至后续的M-1个共享特征提取块，作为后续M-1个共享特征提取块的输入特征。
在本申请一些实施例中，第二个共享特征提取块的输入特征即为第一个共享特征提取块的输出特征，第二个共享特征提取块对其输入特征进行特征提取处理后，可将其输出特征输出至后续的第3个至第M个共享特征提取块，作为第3个至第M个共享特征提取块的输入特征。
在本申请一些实施例中，第3个共享特征提取块的输入特征为第一个共享特征提取块的输出特征和第二个共享特征提取块的输出特征，第一个共享特征提取块的输出特征和第二个共享特征提取块的输出特征可在特征融合(例如，通过计算平均值、最大值等方式进行融合，或者保留所有特征通道)后输入至第3个共享特征提取块(即，第3个共享特征提取块的输入特征可以是第一个共享特征提取块的输出特征和第二个共享特征提取块的输出特征融合后的特征)，或者，第3个共享特征提取块可直接将第一个共享特征提取块的输出特征和第二个共享特征提取块的输出特征均作为输入特征(例如，第3个共享特征提取块可包括特征融合层，可由该层级进行特征融合处理，或者保留所有特征通道，并可直接对所有特征通道的特征进行后续处理，即，对第一个共享特征提取块的输出特征和第二个共享特征提取块的输出特征进行后续处理)，并对输入特征进行特征提取处理(例如，直接对所有特征通道的特征进行特征提取处理，或者对融合后的特征进行特征提取处理)，获得第3个共享特征提取块的输出特征。第3个共享特征提取块的输出特征可输出至第4个至第M个共享特征提取块，作为第4个至第M个共享特征提取块的输入特征。
在本申请一些实施例中，以第j(j为整数且1＜j＜M)个共享特征提取块为例，前j-1个共享特征提取块的输出特征可被作为输入特征输入至第j个共享特征提取块。可将前j-1个共享特征提取块的输出特征进行特征融合后，将融合后的特征作为第j个共享特征提取块的输入特征，或者直接将前j-1个共享特征提取块的输出特征作为第j个共享特征提取块的输入特征(例如，在第j个共享特征提取块内进行融合，或者直接对所有特征通道的特征进行后续处理，即，将前j-1个共享特征提取块的输出特征进行后续处理)。第j个共享特征提取块可对其输入特征进行特征提取处理，获得第j个共享特征提取块的输出特征，并将该输出特征作为第j+1个至第M个共享特征提取块的输入特征。
在本申请一些实施例中，第M个共享特征提取块可根据前M-1个共享特征提取块的输出特征，获得第M个共享特征提取块的输出特征。并可通过共享特征提取网络的后续的网络层级进行第二特征提取处理，例如，可通过包括三维卷积层、批归一化层和激活层的网络模块对第N个共享特征提取块的输出特征进行第二特征提取处理，获得第二特征。本申请实施例对进行第二特征提取处理的网络层级不做限制。
在本申请一些实施例中，可对第二特征进行池化处理，例如，可通过平均值池化层对第二特征进行池化处理获得目标特征。本申请实施例对池化处理的类型不做限制。
在本申请一些实施例中，上述处理可进行多次，例如，可包括多个共享特征提取网络。第一个共享特征提取网络可以以第一特征为输入特征，经过共享特征提取块的特征提取处理、第二特征提取处理和池化处理后，获得第一个共享特征提取网络的输出特征，第二个共享特征提取网络可以将第一个共享特征提取网络的输出特征作为输入特征，经过共享特征提取块的特征提取处理、第二特征提取处理和池化处理后，获得第二个共享特征提取网络的输出特征……可通过多个共享特征提取网络进行处理，并将最后一个(例如，第4个)共享特征提取网络的输出特征作为目标特征。本申请实施例对共享特征提取网络数量不做限制。
通过这种方式，能够通过共享特征提取网络来获得目标特征，共享特征提取网络的共享特征提取块可获得之前所有共享特征提取块的输出特征，并将自身的输出特征输入至后续所有共享特征提取块。可加强网络内的梯度流动，缓解梯度消失现象，同时提高 特征提取和学习能力，有利于对输入的待处理图像块进行更精细分类和分割处理。
在本申请一些实施例中，可根据目标特征来确定待处理图像块的类别信息，例如，待处理图像块是肺部三维医学图像中包括结节等病灶的图像块，可根据目标特征确定结节的类别。在示例中，可确定结节的类别为浸润前腺癌非典型腺瘤增生、原位腺癌、微创腺癌还是浸润性腺癌。
在本申请一些实施例中，可通过分类网络对目标特征进行分类处理，获得待处理图像块的类别信息。在示例中，分类网络可包括多个网络层级，例如，卷积层、全局平均池化层(Global Average Pooling)和全连接层(Fully Connected Layer)等，上述网络层级可对目标特征进行分类处理，并可输出类别信息。所述类别信息可以是通过向量等形式表示的类别信息，可通过概率词典等确定该向量表示的待处理图像块属于各类别的概率分布，进而确定待处理图像块的类别信息。或者，类别信息的向量可直接表示待处理图像块的概率，在示例中，该向量的各元素分别表示待处理图像块所属类别的概率，例如，(0.8、0.1、0.1)可表示待处理图像块属于第一个类别的概率为0.8，属于第二个类别的概率为0.1，属于第三个类别的概率为0.1，并可将概率最大的类别确定为待处理图像块的类别，即，可将待处理图像块的类别信息确定为第一个类别。本申请实施例对类别信息的表示方法不做限制。
在本申请一些实施例中，可根据目标特征来确定待处理图像块的类别信息，例如，待处理图像块是肺部三维医学图像中包括结节等病灶的图像块，可根据目标特征确定结节的位置，并对其所在区域进行分割。
在本申请一些实施例中，可通过分割网络进行分割处理，获得待处理图像块中的目标区域，例如，可分割出目标区域。在示例中，分割网络可包括多个网络层级，例如，上采样层(Upsample)、全连接层等。在示例中，目标特征为共享特征提取网络中经过对待处理图像块进行特征提取、池化等处理获得的特征图，目标特征的分辨率可低于待处理图像块。可通过上采样层进行上采样处理，将目标特征的特征通道数量减少，并提升分辨率，使分割网络输出的特征图与待处理图像块分辨率一致。例如，共享特征提取网络进行四次池化处理，则可通过上采样层进行四次上采样处理，以使分割网络的输出的特征图与待处理图像块分辨率一致。并可在分割网络的输出的特征图中分割出目标区域，例如，通过轮廓线或轮廓面确定出结节所在的目标区域。本申请实施例对分割网络的网络层级不做限制。
在本申请一些实施例中，在待处理图像块中分割出目标区域后，还可确定目标区域在待处理图像中的位置。例如，可根据待处理图像块在待处理图像中的位置，以及待处理图像块中目标区域的位置，还原出目标区域在待处理图像中的位置。在示例中，在肺部医学图像中，可分割出待处理图像块中结节所在位置，并还原出该结节在肺部医学图像中的位置。
根据本申请实施例提供的神经网络训练方法，可获得样本图像块的精细分类，并对神经网络进行训练，使得神经网络可对图像进行精细分类，提高分类效率和准确度；并可通过共享特征提取网络来获得目标特征，共享特征提取网络的共享特征提取块可获得之前所有共享特征提取块的输出特征，将自身的输出特征输入至后续所有共享特征提取块，以加强网络内的梯度流动，缓解梯度消失现象，同时提高特征提取和学习能力，有利于对输入的待处理图像块进行更精细分类和分割处理；并可获得待处理图像块的较精细的类别信息和目标区域，提升图像处理效率；并可在训练中使相同类别的样本图像块的类别信息更聚集，使不同类别的样本图像块的类别信息之间的特征距离更大；并可平衡目标区域的误差和非目标区域的误差，有助于提升分类性能，提高分类准确率。
图3是本申请实施例提供的神经网络训练方法的一种应用示意图，如图3所示，样 本图像31为医学影像图片，样本图像块32为医学影像图片中剪裁出的包括病灶(例如，结节)的图像块。并且，样本图像块可具有类别标注，例如，样本图像块可包括AAHOPA、AIS、MIA和IA四个类别。
在本申请一些实施例中，可将样本图像块32输入神经网络33，神经网络33包括的共享特征提取网络331对每批样本图像块进行特征提取，获得样本图像块的样本目标特征，并通过神经网络33包括的分类网络332获得样本图像块的类别预测信息，通过公式(1)和公式(2)，可确定神经网络的分类损失。在本申请一些实施例中，神经网络33包括的分割网络333可获得样本图像块32中的预测目标区域，并可根据公式(3)确定神经网络的分割损失。可对分割损失和分类损失加权求和，获得神经网络的综合网络损失，并通过综合网络损失训练神经网络。训练后的神经网络可用于确定医学影像的图像块中的病灶区域和病灶类别。
在本申请一些实施例中，待处理图像可以是三维肺部医学图像(例如，肺部CT图像)，待处理图像块可以是待处理图像中剪裁出的病例区域(例如，具有结节的区域)的三维图像块。
在本申请一些实施例中，可对三维医学图像进行重采样处理，获得分辨率为1×1×1的三维图像，并剪裁出肺部所在区域，进而可对肺部所在区域进行归一化。在本申请一些实施例中，可检测肺部所在区域中的结节所在区域，并按照64×64×64的尺寸剪裁出包括结节所在区域的多个待处理图像块。
在本申请一些实施例中，可将多个待处理图像块分批进行特征提取处理，获得待处理图像块的目标特征。例如，可首先进行第一特征提取处理，例如，可通过包括三维卷积层、批归一化层和激活层的网络模块来进行第一特征提取处理，获得第一特征。
在本申请一些实施例中，可将第一特征输入共享特征提取网络。共享特征提取网络可包括多个共享特征提取块。在示例中，共享特征提取块的数量为M个，可将第一特征输入第一个共享特征提取块进行处理，第一个共享特征提取块的输出特征可至后续的M-1个共享特征提取块。第二个共享特征提取块的输入特征即为第一个共享特征提取块的输出特征，并且，第二个共享特征提取块可将其输出特征输出至后续的第3个至第M个共享特征提取块。第3个共享特征提取块的输入特征为第一个共享特征提取块的输出特征和第二个共享特征提取块的输出特征，并且，第3个共享特征提取块的输出特征可输出至第4个至第M个共享特征提取块。类似地，前j-1个共享特征提取块的输出特征可被输入至第j个共享特征提取块，第j个共享特征提取块的输出特征可输出至第j+1个至第M个共享特征提取块。第M个共享特征提取块可根据前M-1个共享特征提取块的输出特征，获得第M个共享特征提取块的输出特征，并进行第二特征提取处理，例如，可通过包括三维卷积层、批归一化层和激活层的网络模块对第N个共享特征提取块的输出特征进行第二特征提取处理，获得第二特征。在本申请一些实施例中，可对第二特征进行池化(例如，平均值池化(Average Pooling))处理，获得目标特征。
在本申请一些实施例中，上述处理可进行多次(例如4次)，例如，可包括多个共享特征提取网络。经过多个级联的共享特征提取网络的处理，可获得目标特征。
在本申请一些实施例中，分类网络可对目标特征进行分类处理，获得待处理图像块的类别信息。例如，分类网络可通过卷积层、全局平均池化层和全连接层等，获得待处理图像块的类别信息。
在本申请一些实施例中，分割网络可对目标特征进行分割处理，获得目标区域(即，结节所在区域)。在示例中，分割网络通过上采样层进行四次上采样处理，以使分割网络的输出的特征图与待处理图像块分辨率一致，并可在分割网络的输出的特征图中分割出目标区域。
在本申请一些实施例中，上述神经网络可在待处理图像块中目标区域和类别均未知的情况下，获得待处理图像块的类别和目标区域(例如，可分割出结节所在区域，并获得结节的类别)，也可在待处理图像块的类别已知的情况下，仅获取待处理图像块中的目标区域(例如，分割出结节所在区域)，或者可在待处理图像块中目标区域已知的情况下，获取待处理图像块的类别(例如，确定结节的类别)。
在本申请一些实施例中，所述图像处理方法可用于对肺部CT图像等医学图像中的病例区域进行分割和分类，提高临床工作效率，减少漏诊和误诊，也可用于对其他图像进行分类和目标区域的分割，本申请实施例对所述图像处理方法的应用领域不做限制。
可以理解，本申请提及的上述各个方法实施例，在不违背原理逻辑的情况下，均可以彼此相互结合形成结合后的实施例。此外，本申请实施例还提供了装置、电子设备、计算机可读存储介质、程序，上述均可用来实现本申请实施例提供的任一种方法，相应技术方案和描述和参见方法部分的相应记载。本领域技术人员可以理解，在上述方法实施例中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的执行顺序应当以其功能和可能的内在逻辑确定。
图4是本申请实施例提供的神经网络训练装置的一种示意图，如图4所示，所述装置包括：获取模块11，配置为获取样本图像中目标区域的位置信息及类别信息；第一分割模块12，配置为根据所述样本图像中目标区域的位置信息，分割得到至少一个样本图像块；分类模块13，配置为根据所述类别信息，将所述至少一个样本图像块进行分类，得到N类样本图像块，N为整数，且N≥1；训练模块14，配置为将所述N类样本图像块输入至神经网络中进行训练。
在本申请一些实施例中，所述样本图像为医学影像图片。
在本申请一些实施例中，所述获取模块11还配置为：对医学影像图片上的目标区域进行定位，得到所述目标区域的位置信息；获取与所述医学影像图片关联的病理学图片；所述病理学图片为经过诊断的包括病理信息的图片；根据所述病理学图片上的各目标区域的病理信息，确定所述医学影像图片上的目标区域的类别信息。
在本申请一些实施例中，所述训练模块14还配置为：将任一的样本图像块输入所述神经网络进行处理，获得样本图像块的类别预测信息和预测目标区域；至少根据所述类别预测信息和所述样本图像块的类别信息，确定分类损失；根据所述预测目标区域和所述样本图像块的位置信息，确定分割损失；根据所述分类损失和所述分割损失，训练所述神经网络。
在本申请一些实施例中，所述训练模块14还配置为：根据所述类别预测信息和所述样本图像块的类别信息，确定第一分类损失；根据所述类别预测信息和所述样本图像块所属类别的类中心的类别信息，确定第二分类损失；对所述第一分类损失和所述第二分类损失进行加权求和处理，获得所述分类损失。
在本申请一些实施例中，所述训练模块14还配置为：根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中样本背景区域的第二权重；根据所述第一权重、第二权重、所述预测目标区域和所述样本图像块的位置信息，确定所述分割损失。
在本申请一些实施例中，所述训练模块14还配置为：根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述样本图像块中样本背景区域的第二比例；将所述第二比例确定为所述第一权重，并将所述第一比例确定为第二权重。
在本申请一些实施例中，所述类别信息包括：浸润前腺癌非典型腺瘤增生结节、原位腺癌结节、微创腺癌结节和浸润性腺癌结节。
在本申请一些实施例中，所述神经网络包括共享特征提取网络、分类网络和分割网 络，所述装置还包括：获得模块，配置为将待处理图像块输入所述共享特征提取网络进行处理，获得待处理图像块的目标特征，其中，所述共享特征提取网络包括M个共享特征提取块，第i个共享特征提取块的输入特征包括前i-1个共享特征提取块的输出特征，i和M为整数且1＜i≤M；分类模块，配置为将所述目标特征输入所述分类网络进行分类处理，获得所述待处理图像块的类别信息；分割模块，配置为将所述目标特征输入所述分割网络进行分割处理，获得所述待处理图像块中的目标区域。
在本申请一些实施例中，所述获得模块还配置为：对所述待处理图像块进行第一特征提取处理，获得所述待处理图像块的第一特征；将所述第一特征输入第一个共享特征提取块，获得所述第一个共享特征提取块的输出特征，并将所述第一个共享特征提取块的输出特征输出至后续的M-1个共享特征提取块；将前j-1个共享特征提取块的输出特征输入至第j个共享特征提取块，获得所述第j个共享特征提取块的输出特征，其中，j为整数且1＜j＜M；将第M个共享特征提取块的输出特征进行第二特征提取处理，获得所述待处理图像块的第二特征；对所述第二特征进行池化处理，获得所述目标特征。
在本申请一些实施例中，所述装置还包括：预处理模块，配置为对待处理图像进行预处理，获得第一图像；定位模块，配置为对所述第一图像上的目标区域进行定位，确定所述第一图像中的目标区域的位置信息；第二分割模块，配置为根据所述第一图像中的目标区域的位置信息，分割得到至少一个所述待处理图像块。
在一些实施例中，本申请实施例提供的装置具有的功能或包含的模块可以配置为执行上文方法实施例描述的方法，其实现可以参照上文方法实施例的描述。
本申请实施例还提供一种计算机可读存储介质，其上存储有计算机程序，所述计算机程序被配置为运行时执行上述方法。计算机可读存储介质可以是非易失性计算机可读存储介质。
本申请实施例还提供一种电子设备，包括：处理器；配置为存储处理器可执行计算机程序的存储器；其中，所述处理器被配置为通过所述计算机程序执行上述方法。电子设备可以被提供为终端、服务器或其它形态的设备。
本申请实施例还提供一种计算机程序产品，包括计算机可读代码，当计算机可读代码在设备上运行时，设备中的处理器执行如上所述任一实施例提供的神经网络训练方法的指令。
本申请实施例还提供另一种计算机程序产品，配置为存储计算机可读指令，指令被执行时使得计算机执行上述任一实施例提供的神经网络训练方法的操作。
图5是本申请实施例提供的一种电子设备的示意图。例如，电子设备800可以是移动电话，计算机，数字广播终端，消息收发设备，游戏控制台，平板设备，医疗设备，健身设备，个人数字助理等终端。参照图5，电子设备800可以包括以下一个或多个组件：处理组件802，存储器804，电源组件806，多媒体组件808，音频组件810，输入/输出(Input/Output，I/O)的接口812，传感器组件814，以及通信组件816。
处理组件802通常控制电子设备800的整体操作，诸如与显示，电话呼叫，数据通信，相机操作和记录操作相关联的操作。处理组件802可以包括一个或多个处理器820来执行指令，以完成上述的方法的全部或部分步骤。此外，处理组件802可以包括一个或多个模块，便于处理组件802和其他组件之间的交互。例如，处理组件802可以包括多媒体模块，以方便多媒体组件808和处理组件802之间的交互。
存储器804被配置为存储各种类型的数据以支持在电子设备800的操作。这些数据的示例包括用于在电子设备800上操作的任何应用程序或方法的指令，联系人数据，电话簿数据，消息，图片，视频等。存储器804可以由任何类型的易失性或非易失性存储设备或者它们的组合实现，如静态随机存取存储器(Static Random Access Memory， SRAM)，电可擦除可编程只读存储器(Electrically Erasable Programmable Read Only Memory，EEPROM)，可擦除可编程只读存储器(Erasable Programmable Read Only Memory，EPROM)，可编程只读存储器(Programmable Read Only Memory，PROM)，只读存储器(Read Only Memory，ROM)，磁存储器，快闪存储器，磁盘或光盘。
电源组件806为电子设备800的各种组件提供电力，可包括电源管理系统，一个或多个电源，及其他与为电子设备800生成、管理和分配电力相关联的组件。
多媒体组件808包括在所述电子设备800和用户之间的提供一个输出接口的屏幕。在一些实施例中，屏幕可以包括液晶显示器(Liquid Crystal Display，LCD)和触摸面板(TouchPanel，TP)。如果屏幕包括触摸面板，屏幕可以被实现为触摸屏，以接收来自用户的输入信号。触摸面板包括一个或多个触摸传感器以感测触摸、滑动和触摸面板上的手势。所述触摸传感器可以不仅感测触摸或滑动动作的边界，而且还检测与所述触摸或滑动操作相关的持续时间和压力。在一些实施例中，多媒体组件808包括一个前置摄像头和/或后置摄像头。当电子设备800处于操作模式，如拍摄模式或视频模式时，前置摄像头和/或后置摄像头可以接收外部的多媒体数据。每个前置摄像头和后置摄像头可以是一个固定的光学透镜系统或具有焦距和光学变焦能力。
音频组件810被配置为输出和/或输入音频信号。例如，音频组件810包括一个麦克风(Microphone，MIC)，当电子设备800处于操作模式，如呼叫模式、记录模式和语音识别模式时，麦克风被配置为接收外部音频信号。所接收的音频信号可以被存储在存储器804或经由通信组件816发送。在一些实施例中，音频组件810还包括一个扬声器，配置为输出音频信号。
I/O接口812为处理组件802和外围接口模块之间提供接口，上述外围接口模块可以是键盘，点击轮，按钮等。这些按钮可包括但不限于：主页按钮、音量按钮、启动按钮和锁定按钮。
传感器组件814包括一个或多个传感器，配置为为电子设备800提供各个方面的状态评估。例如，传感器组件814可以检测到电子设备800的打开/关闭状态，组件的相对定位，例如所述组件为电子设备800的显示器和小键盘，传感器组件814还可以检测电子设备800或电子设备800一个组件的位置改变，用户与电子设备800接触的存在或不存在，电子设备800方位或加速/减速和电子设备800的温度变化。传感器组件814可以包括接近传感器，被配置用来在没有任何的物理接触时检测附近物体的存在。传感器组件814还可以包括光传感器，如互补金属氧化物半导体(Complementary Metal Oxide Semiconductor，CMOS)或电荷耦合器件(Charge Coupled Device，CCD)图像传感器，可在成像应用中使用。在一些实施例中，该传感器组件814还可以包括加速度传感器，陀螺仪传感器，磁传感器，压力传感器或温度传感器。
通信组件816被配置为便于电子设备800和其他设备之间有线或无线方式的通信。电子设备800可以接入基于通信标准的无线网络，如WiFi，2G或3G，或它们的组合。在本申请一些实施例中，通信组件816经由广播信道接收来自外部广播管理系统的广播信号或广播相关信息。在本申请一些实施例中，所述通信组件816还包括近场通信(Near Field Communication，NFC)模块，以促进短程通信。例如，在NFC模块可基于射频识别(Radio Frequency Identification，RFID)技术，红外数据协会(Infrared Data Association，IrDA)技术，超宽带(Ultra Wide Band，UWB)技术，蓝牙(Bluetooth，BT)技术和其他技术来实现。
在本申请一些实施例中，电子设备800可以被一个或多个应用专用集成电路(Application Specific Integrated Circuit，ASIC)、数字信号处理器(Digital Signal Process，DSP)、数字信号处理设备(Digital Signal Process Device，DSPD)、可编程逻辑器件 (Programmable Logic Device，PLD)、现场可编程门阵列(Field Programmable Gate Array，FPGA)、控制器、微控制器、微处理器或其他电子元件实现，配置为执行上述方法。
在本申请一些实施例中，还提供一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器804，上述计算机程序指令可由电子设备800的处理器820执行以完成上述方法。
图6是本申请实施例提供的另一种电子设备的示意图。例如，电子设备1900可以被提供为一服务器。参照图6，电子设备1900包括处理组件1922，还可以包括一个或多个处理器，以及由存储器1932所代表的存储器资源，配置为存储可由处理组件1922的执行的指令，例如应用程序。存储器1932中存储的应用程序可以包括一个或一个以上的每一个对应于一组指令的模块。此外，处理组件1922被配置为执行指令，以执行上述方法。
电子设备1900还可以包括一个电源组件1926被配置为执行电子设备1900的电源管理，一个有线或无线网络接口1950被配置为将电子设备1900连接到网络，和一个输入输出(I/O)接口1958。电子设备1900可以操作基于存储在存储器1932的操作系统，例如Windows ServerTM，Mac OS XTM，UnixTM，LinuxTM，FreeBSDTM或类似。
在本申请一些实施例中，还提供一种非易失性计算机可读存储介质，例如包括计算机程序指令的存储器1932，上述计算机程序指令可由电子设备1900的处理组件1922执行以完成上述方法。
本申请可以是系统、方法和/或计算机程序产品。计算机程序产品可包括计算机可读存储介质，其上载有配置为使处理器实现本申请的各个方面的计算机可读程序指令。
计算机可读存储介质可以是保持和存储由指令执行设备使用的指令的有形设备。计算机可读存储介质例如可以是但不限于：电存储设备、磁存储设备、光存储设备、电磁存储设备、半导体存储设备或者上述的任意合适的组合。计算机可读存储介质可以包括：便携式计算机盘、硬盘、随机存取存储器(Random Access Memory，RAM)、ROM、可擦式可编程只读存储器(EPROM或闪存)、SRAM、便携式压缩盘只读存储器(Compact Disk-Read Only Memory，CD-ROM)、数字多功能盘(Digital Video Disc，DVD)、记忆棒、软盘、机械编码设备、例如其上存储有指令的打孔卡或凹槽内凸起结构、以及上述的任意合适的组合。这里所使用的计算机可读存储介质不被解释为瞬时信号本身，诸如无线电波或者其他自由传播的电磁波、通过波导或其他传输媒介传播的电磁波(如，通过光纤电缆的光脉冲)、或者通过电线传输的电信号。
这里所描述的计算机可读程序指令可以从计算机可读存储介质下载到各个计算/处理设备，或者通过网络、例如因特网、局域网、广域网和/或无线网下载到外部计算机或外部存储设备。网络可以包括铜传输电缆、光纤传输、无线传输、路由器、防火墙、交换机、网关计算机和/或边缘服务器。每个计算/处理设备中的网络适配卡或者网络接口从网络接收计算机可读程序指令，并转发该计算机可读程序指令，以供存储在各个计算/处理设备中的计算机可读存储介质中。
配置为执行本申请操作的计算机程序指令可以是汇编指令、指令集架构(Instruction Set Architecture，ISA)指令、机器指令、机器相关指令、微代码、固件指令、状态设置数据、或者以一种或多种编程语言的任意组合编写的源代码或目标代码，所述编程语言包括面向对象的编程语言，诸如Smalltalk、C++等，以及常规的过程式编程语言，诸如“C”语言或类似的编程语言。计算机可读程序指令可以完全地在用户计算机上执行、部分地在用户计算机上执行、作为一个独立的软件包执行、部分在用户计算机上部分在远程计算机上执行、或者完全在远程计算机或服务器上执行。在涉及远程计算机的情形 中，远程计算机可以通过任意种类的网络(包括局域网或广域网)连接到用户计算机，或者，可以连接到外部计算机(例如利用因特网服务提供商来通过因特网连接)。在一些实施例中，通过利用计算机可读程序指令的状态信息来个性化定制电子电路，例如可编程逻辑电路、FPGA或可编程逻辑阵列(Programmable Logic Array，PLA)，该电子电路可以执行计算机可读程序指令，从而实现本申请的各个方面。
这里参照根据本申请实施例的方法、装置(系统)和计算机程序产品的流程图和/或框图描述了本申请的各个方面。应当理解，流程图和/或框图的每个方框以及流程图和/或框图中各方框的组合，都可以由计算机可读程序指令实现。
这些计算机可读程序指令可以提供给通用计算机、专用计算机或其它可编程数据处理装置的处理器，从而生产出一种机器，使得这些指令在通过计算机或其它可编程数据处理装置的处理器执行时，产生了实现流程图和/或框图中的一个或多个方框中规定的功能/动作的装置。也可以把这些计算机可读程序指令存储在计算机可读存储介质中，这些指令使得计算机、可编程数据处理装置和/或其他设备以特定方式工作，从而，存储有指令的计算机可读介质则包括一个制造品，其包括实现流程图和/或框图中的一个或多个方框中规定的功能/动作的各个方面的指令。
也可以把计算机可读程序指令加载到计算机、其它可编程数据处理装置、或其它设备上，使得在计算机、其它可编程数据处理装置或其它设备上执行一系列操作步骤，以产生计算机实现的过程，从而使得在计算机、其它可编程数据处理装置、或其它设备上执行的指令实现流程图和/或框图中的一个或多个方框中规定的功能/动作。
附图中的流程图和框图显示了根据本申请的多个实施例的系统、方法和计算机程序产品的可能实现的体系架构、功能和操作。在这点上，流程图或框图中的每个方框可以代表一个模块、程序段或指令的一部分，所述模块、程序段或指令的一部分包含一个或多个配置为实现规定的逻辑功能的可执行指令。在有些作为替换的实现中，方框中所标注的功能也可以以不同于附图中所标注的顺序发生。例如，两个连续的方框实际上可以基本并行地执行，它们有时也可以按相反的顺序执行，这依所涉及的功能而定。也要注意的是，框图和/或流程图中的每个方框、以及框图和/或流程图中的方框的组合，可以用执行规定的功能或动作的专用的基于硬件的系统来实现，或者可以用专用硬件与计算机指令的组合来实现。
该计算机程序产品可以通过硬件、软件或其结合的方式实现。在本申请实施例一实施方式中，所述计算机程序产品可以体现为计算机存储介质，在另一实施方式中，计算机程序产品可以体现为软件产品，例如软件开发包(Software Development Kit，SDK)等等。
以上已经描述了本申请的各实施例，上述说明是示例性的，并非穷尽性的，并且也不限于所披露的各实施例。在不偏离所说明的各实施例的范围和精神的情况下，对于本技术领域的普通技术人员来说许多修改和变更都是显而易见的。本文中所用术语的选择，旨在最好地解释各实施例的原理、实际应用或对相关技术的技术改进，或者使本技术领域的其它普通技术人员能理解本文披露的各实施例。
工业实用性
本申请涉及一种神经网络训练方法及装置、电子设备和存储介质，所述方法包括：获取样本图像中目标区域的位置信息及类别信息；根据目标区域的位置信息，分割得到至少一个样本图像块；根据类别信息，将至少一个样本图像块进行分类，得到N类样本图像块；将N类样本图像块输入至神经网络中进行训练。根据本申请的实施例的神经网络训练方法，可获得样本图像块的精细分类，并对神经网络进行训练，使得神经网络可对图像进行精细分类，提高分类效率和准确度。

Claims (25)

  1. 一种神经网络训练方法，所述神经网络训练方法用于训练神经网络模型，根据训练得到的神经网络模型对图像进行分类，所述方法包括：
    获取样本图像中目标区域的位置信息及类别信息；
    根据所述样本图像中目标区域的位置信息，分割得到至少一个样本图像块；
    根据所述类别信息，将所述至少一个样本图像块进行分类，得到N类样本图像块，N为整数，且N≥1；
    将所述N类样本图像块输入至神经网络中进行训练。
  2. 根据权利要求1所述的方法，其中，所述样本图像为医学影像图片。
  3. 根据权利要求2所述的方法，其中，所述获取样本图像中目标区域的位置信息及类别信息包括：
    对所述医学影像图片上的目标区域进行定位，得到所述目标区域的位置信息；
    获取与所述医学影像图片关联的病理学图片，所述病理学图片为经过诊断的包括病理信息的图片；
    根据所述病理学图片上的各目标区域的病理信息，确定所述医学影像图片上的目标区域的类别信息。
  4. 根据权利要求1所述的方法，其中，将所述N类样本图像块输入至神经网络中进行训练，包括：
    将任一的样本图像块输入所述神经网络进行处理，获得样本图像块的类别预测信息和预测目标区域；
    至少根据所述类别预测信息和所述样本图像块的类别信息，确定分类损失；
    根据所述预测目标区域和所述样本图像块的位置信息，确定分割损失；
    根据所述分类损失和所述分割损失，训练所述神经网络。
  5. 根据权利要求4所述的方法，其中，根据所述类别预测信息和所述样本图像块的类别信息，确定分类损失，包括：
    根据所述类别预测信息和所述样本图像块的类别信息，确定第一分类损失；
    根据所述类别预测信息和所述样本图像块所属类别的类中心的类别信息，确定第二分类损失；
    对所述第一分类损失和所述第二分类损失进行加权求和处理，获得所述分类损失。
  6. 根据权利要求4所述的方法，其中，根据所述预测目标区域和所述样本图像块的位置信息，确定分割损失，包括：
    根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中样本背景区域的第二权重；
    根据所述第一权重、第二权重、所述预测目标区域和所述样本图像块的位置信息，确定所述分割损失。
  7. 根据权利要求6所述的方法，其中，根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中 样本背景区域的第二权重，包括：
    根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述样本图像块中样本背景区域的第二比例；
    将所述第二比例确定为所述第一权重，并将所述第一比例确定为第二权重。
  8. 根据权利要求1至7任意一项所述的方法，其中，所述类别信息包括：浸润前腺癌非典型腺瘤增生结节、原位腺癌结节、微创腺癌结节和浸润性腺癌结节。
  9. 根据权利要求1至8任意一项所述的方法，其中，所述神经网络包括共享特征提取网络、分类网络和分割网络；
    所述方法还包括：
    将待处理图像块输入所述共享特征提取网络进行处理，获得所述待处理图像块的目标特征，其中，所述共享特征提取网络包括M个共享特征提取块，第i个共享特征提取块的输入特征包括前i-1个共享特征提取块的输出特征，i和M为整数且1＜i≤M；
    将所述目标特征输入所述分类网络进行分类处理，获得所述待处理图像块的类别信息；
    将所述目标特征输入所述分割网络进行分割处理，获得所述待处理图像块中的目标区域。
  10. 根据权利要求9所述的方法，其中，将待处理图像块输入所述共享特征提取网络进行处理，获得所述待处理图像块的目标特征，包括：
    对所述待处理图像块进行第一特征提取处理，获得所述待处理图像块的第一特征；
    将所述第一特征输入第一个共享特征提取块，获得所述第一个共享特征提取块的输出特征，并将所述第一个共享特征提取块的输出特征输出至后续的M-1个共享特征提取块；
    将前j-1个共享特征提取块的输出特征输入至第j个共享特征提取块，获得所述第j个共享特征提取块的输出特征，其中，j为整数且1＜j＜M；
    将第M个共享特征提取块的输出特征进行第二特征提取处理，获得所述待处理图像块的第二特征；
    对所述第二特征进行池化处理，获得所述目标特征。
  11. 根据权利要求9所述的方法，其中，所述方法还包括：
    对待处理图像进行预处理，获得第一图像；
    对所述第一图像上的目标区域进行定位，确定所述第一图像中的目标区域的位置信息；
    根据所述第一图像中的目标区域的位置信息，分割得到至少一个所述待处理图像块。
  12. 一种神经网络训练装置，所述神经网络训练装置用于训练神经网络模型，根据训练得到的神经网络模型对图像进行分类，所述装置包括：
    获取模块，配置为获取样本图像中目标区域的位置信息及类别信息；
    第一分割模块，配置为根据所述样本图像中目标区域的位置信息，分割得到至少一个样本图像块；
    分类模块，配置为根据所述类别信息，将所述至少一个样本图像块进行分类，得到 N类样本图像块，N为整数，且N≥1；
    训练模块，配置为将所述N类样本图像块输入至神经网络中进行训练。
  13. 根据权利要求12所述的装置，其中，所述样本图像为医学影像图片。
  14. 根据权利要求13所述的装置，其中，所述获取模块还配置为：
    对所述医学影像图片上的目标区域进行定位，得到所述目标区域的位置信息；
    获取与所述医学影像图片关联的病理学图片，所述病理学图片为经过诊断的包括病理信息的图片；
    根据所述病理学图片上的各目标区域的病理信息，确定所述医学影像图片上的目标区域的类别信息。
  15. 根据权利要求12所述的装置，其中，所述训练模块还配置为：
    将任一的样本图像块输入所述神经网络进行处理，获得样本图像块的类别预测信息和预测目标区域；
    至少根据所述类别预测信息和所述样本图像块的类别信息，确定分类损失；
    根据所述预测目标区域和所述样本图像块的位置信息，确定分割损失；
    根据所述分类损失和所述分割损失，训练所述神经网络。
  16. 根据权利要求15所述的装置，其中，所述训练模块还配置为：
    根据所述类别预测信息和所述样本图像块的类别信息，确定第一分类损失；
    根据所述类别预测信息和所述样本图像块所属类别的类中心的类别信息，确定第二分类损失；
    对所述第一分类损失和所述第二分类损失进行加权求和处理，获得所述分类损失。
  17. 根据权利要求15所述的装置，其中，所述训练模块还配置为：
    根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述预测目标区域的第一权重和所述样本图像块中样本背景区域的第二权重；
    根据所述第一权重、第二权重、所述预测目标区域和所述样本图像块的位置信息，确定所述分割损失。
  18. 根据权利要求17所述的装置，其中，所述训练模块还配置为：
    根据所述预测目标区域的像素数量在所述样本图像块中所占的第一比例，确定所述样本图像块中样本背景区域的第二比例；
    将所述第二比例确定为所述第一权重，并将所述第一比例确定为第二权重。
  19. 根据权利要求12至18任意一项所述的装置，其中，所述类别信息包括：浸润前腺癌非典型腺瘤增生结节、原位腺癌结节、微创腺癌结节和浸润性腺癌结节。
  20. 根据权利要求12至19任意一项所述的装置，其中，所述神经网络包括共享特征提取网络、分类网络和分割网络；
    所述装置还包括：
    获得模块，配置为将待处理图像块输入所述共享特征提取网络进行处理，获得所述待处理图像块的目标特征，其中，所述共享特征提取网络包括M个共享特征提取块，第i个共享特征提取块的输入特征包括前i-1个共享特征提取块的输出特征，i和M为整数且1＜i≤M；
    分类模块，配置为将所述目标特征输入所述分类网络进行分类处理，获得所述待处 理图像块的类别信息；
    分割模块，配置为将所述目标特征输入所述分割网络进行分割处理，获得所述待处理图像块中的目标区域。
  21. 根据权利要求20所述的装置，其中，所述获得模块还配置为：
    对所述待处理图像块进行第一特征提取处理，获得所述待处理图像块的第一特征；
    将所述第一特征输入第一个共享特征提取块，获得所述第一个共享特征提取块的输出特征，并将所述第一个共享特征提取块的输出特征输出至后续的M-1个共享特征提取块；
    将前j-1个共享特征提取块的输出特征输入至第j个共享特征提取块，获得所述第j个共享特征提取块的输出特征，其中，j为整数且1＜j＜M；
    将第M个共享特征提取块的输出特征进行第二特征提取处理，获得所述待处理图像块的第二特征；
    对所述第二特征进行池化处理，获得所述目标特征。
  22. 根据权利要求20所述的装置，其中，所述装置还包括：
    预处理模块，配置为对待处理图像进行预处理，获得第一图像；
    定位模块，配置为对所述第一图像上的目标区域进行定位，确定所述第一图像中的目标区域的位置信息；
    第二分割模块，配置为根据所述第一图像中的目标区域的位置信息，分割得到至少一个所述待处理图像块。
  23. 一种电子设备，包括：
    处理器；
    配置为存储处理器可执行计算机程序的存储器；
    其中，所述处理器被配置为：通过所述计算机程序执行权利要求1至11中任意一项所述的方法。
  24. 一种计算机可读存储介质，所述计算机可读存储介质中存储有计算机程序，所述计算机程序被配置为运行时执行权利要求1至11中任意一项所述的方法。
  25. 一种计算机程序，包括计算机可读代码，当所述计算机可读代码在电子设备中运行时，所述电子设备中的处理器执行被配置为实现权利要求1至11任意一项所述的方法。
PCT/CN2020/100715 2020-03-05 2020-07-07 神经网络训练方法及装置、电子设备和存储介质 WO2021174739A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
JP2021574781A JP2022537974A (ja) 2020-03-05 2020-07-07 ニューラルネットワーク訓練方法及び装置、電子機器並びに記憶媒体
KR1020217041454A KR20220009451A (ko) 2020-03-05 2020-07-07 신경망 훈련 방법 및 장치, 전자 기기 및 저장 매체

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010148544.8 2020-03-05
CN202010148544.8A CN111368923B (zh) 2020-03-05 2020-03-05 神经网络训练方法及装置、电子设备和存储介质

Publications (1)

Publication Number Publication Date
WO2021174739A1 true WO2021174739A1 (zh) 2021-09-10

Family

ID=71208701

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/100715 WO2021174739A1 (zh) 2020-03-05 2020-07-07 神经网络训练方法及装置、电子设备和存储介质

Country Status (5)

Country Link
JP (1) JP2022537974A (zh)
KR (1) KR20220009451A (zh)
CN (1) CN111368923B (zh)
TW (1) TWI770754B (zh)
WO (1) WO2021174739A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113793323A (zh) * 2021-09-16 2021-12-14 云从科技集团股份有限公司 一种元器件检测方法、系统、设备及介质
CN113989845A (zh) * 2021-10-29 2022-01-28 北京百度网讯科技有限公司 姿态分类方法和姿态分类模型的训练方法、装置
CN113989407A (zh) * 2021-12-30 2022-01-28 青岛美迪康数字工程有限公司 Ct影像中肢体部位识别模型训练方法及系统
CN113989721A (zh) * 2021-10-29 2022-01-28 北京百度网讯科技有限公司 目标检测方法和目标检测模型的训练方法、装置
CN114037925A (zh) * 2021-09-27 2022-02-11 北京百度网讯科技有限公司 目标检测模型的训练、检测方法、装置及电子设备
US20220084677A1 (en) * 2020-09-14 2022-03-17 Novocura Tech Health Services Private Limited System and method for generating differential diagnosis in a healthcare environment
CN114612824A (zh) * 2022-03-09 2022-06-10 清华大学 目标识别方法及装置、电子设备和存储介质
CN116077066A (zh) * 2023-02-10 2023-05-09 北京安芯测科技有限公司 心电信号分类模型的训练方法、装置及电子设备

Families Citing this family (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111368923B (zh) * 2020-03-05 2023-12-19 上海商汤智能科技有限公司 神经网络训练方法及装置、电子设备和存储介质
CN111767708A (zh) * 2020-07-09 2020-10-13 北京猿力未来科技有限公司 解题模型的训练方法及装置、解题公式生成方法及装置
CN112017162B (zh) * 2020-08-10 2022-12-06 上海杏脉信息科技有限公司 病理图像处理方法、装置、存储介质和处理器
CN112241760A (zh) * 2020-08-25 2021-01-19 浙江大学 网络小额贷款服务中的黑中介自动挖掘方法与系统
CN112328398B (zh) * 2020-11-12 2024-09-27 清华大学 任务处理方法及装置、电子设备和存储介质
CN112561893B (zh) * 2020-12-22 2024-09-06 平安银行股份有限公司 图片匹配方法、装置、电子设备及存储介质
CN112785565B (zh) * 2021-01-15 2024-01-05 上海商汤智能科技有限公司 目标检测方法及装置、电子设备和存储介质
CN112749801A (zh) * 2021-01-22 2021-05-04 上海商汤智能科技有限公司 神经网络训练和图像处理方法及装置
CN112907517B (zh) * 2021-01-28 2024-07-19 上海商汤善萃医疗科技有限公司 一种图像处理方法、装置、计算机设备及存储介质
CN112925938A (zh) * 2021-01-28 2021-06-08 上海商汤智能科技有限公司 一种图像标注方法、装置、电子设备及存储介质
US11967084B2 (en) * 2021-03-09 2024-04-23 Ping An Technology (Shenzhen) Co., Ltd. PDAC image segmentation method, electronic device and storage medium
CN113139471A (zh) * 2021-04-25 2021-07-20 上海商汤智能科技有限公司 目标检测方法及装置、电子设备和存储介质
AU2021204563A1 (en) * 2021-06-17 2023-01-19 Sensetime International Pte. Ltd. Target detection methods, apparatuses, electronic devices and computer-readable storage media
CN113702719B (zh) * 2021-08-03 2022-11-29 北京科技大学 一种基于神经网络的宽带近场电磁定位方法及装置
CN113688975A (zh) * 2021-08-24 2021-11-23 北京市商汤科技开发有限公司 神经网络的训练方法、装置、电子设备及存储介质
CN114049315B (zh) * 2021-10-29 2023-04-18 北京长木谷医疗科技有限公司 关节识别方法、电子设备、存储介质及计算机程序产品
CN114332547B (zh) * 2022-03-17 2022-07-08 浙江太美医疗科技股份有限公司 医学目标分类方法和装置、电子设备和存储介质
CN114743055A (zh) * 2022-04-18 2022-07-12 北京理工大学 一种使用分区决策机制提高图像分类准确率的方法
CN114839340A (zh) * 2022-04-27 2022-08-02 芯视界（北京）科技有限公司 水质生物活性检测方法及装置、电子设备和存储介质
KR20240018229A (ko) * 2022-08-02 2024-02-13 김민구 시내퍼 모델을 이용한 자연어 처리 시스템 및 방법

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160034788A1 (en) * 2014-07-30 2016-02-04 Adobe Systems Incorporated Learning image categorization using related attributes
CN107330263A (zh) * 2017-06-26 2017-11-07 成都知识视觉科技有限公司 一种计算机辅助乳腺浸润性导管癌组织学分级的方法
CN108520518A (zh) * 2018-04-10 2018-09-11 复旦大学附属肿瘤医院 一种甲状腺肿瘤超声图像识别方法及其装置
CN109919961A (zh) * 2019-02-22 2019-06-21 北京深睿博联科技有限责任公司 一种用于颅内cta图像中动脉瘤区域的处理方法及装置
CN111368923A (zh) * 2020-03-05 2020-07-03 上海商汤智能科技有限公司 神经网络训练方法及装置、电子设备和存储介质

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP6031921B2 (ja) * 2012-09-28 2016-11-24 ブラザー工業株式会社 画像処理装置およびプログラム
CN108229267B (zh) * 2016-12-29 2020-10-16 北京市商汤科技开发有限公司 对象属性检测、神经网络训练、区域检测方法和装置
US10423861B2 (en) * 2017-10-16 2019-09-24 Illumina, Inc. Deep learning-based techniques for training deep convolutional neural networks
CN108335313A (zh) * 2018-02-26 2018-07-27 阿博茨德（北京）科技有限公司 图像分割方法及装置
CN109285142B (zh) * 2018-08-07 2023-01-06 广州智能装备研究院有限公司 一种头颈部肿瘤检测方法、装置及计算机可读存储介质
CN109447169B (zh) * 2018-11-02 2020-10-27 北京旷视科技有限公司 图像处理方法及其模型的训练方法、装置和电子系统
CN110245657B (zh) * 2019-05-17 2021-08-24 清华大学 病理图像相似性检测方法及检测装置
CN110210535B (zh) * 2019-05-21 2021-09-10 北京市商汤科技开发有限公司 神经网络训练方法及装置以及图像处理方法及装置
CN110705555B (zh) * 2019-09-17 2022-06-14 中山大学 基于fcn的腹部多器官核磁共振图像分割方法、系统及介质
CN110705626A (zh) * 2019-09-26 2020-01-17 北京市商汤科技开发有限公司 一种图像处理方法及装置、电子设备和存储介质
CN110796656A (zh) * 2019-11-01 2020-02-14 上海联影智能医疗科技有限公司 图像检测方法、装置、计算机设备和存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20160034788A1 (en) * 2014-07-30 2016-02-04 Adobe Systems Incorporated Learning image categorization using related attributes
CN107330263A (zh) * 2017-06-26 2017-11-07 成都知识视觉科技有限公司 一种计算机辅助乳腺浸润性导管癌组织学分级的方法
CN108520518A (zh) * 2018-04-10 2018-09-11 复旦大学附属肿瘤医院 一种甲状腺肿瘤超声图像识别方法及其装置
CN109919961A (zh) * 2019-02-22 2019-06-21 北京深睿博联科技有限责任公司 一种用于颅内cta图像中动脉瘤区域的处理方法及装置
CN111368923A (zh) * 2020-03-05 2020-07-03 上海商汤智能科技有限公司 神经网络训练方法及装置、电子设备和存储介质

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220084677A1 (en) * 2020-09-14 2022-03-17 Novocura Tech Health Services Private Limited System and method for generating differential diagnosis in a healthcare environment
CN113793323A (zh) * 2021-09-16 2021-12-14 云从科技集团股份有限公司 一种元器件检测方法、系统、设备及介质
CN114037925A (zh) * 2021-09-27 2022-02-11 北京百度网讯科技有限公司 目标检测模型的训练、检测方法、装置及电子设备
CN113989845A (zh) * 2021-10-29 2022-01-28 北京百度网讯科技有限公司 姿态分类方法和姿态分类模型的训练方法、装置
CN113989721A (zh) * 2021-10-29 2022-01-28 北京百度网讯科技有限公司 目标检测方法和目标检测模型的训练方法、装置
CN113989407A (zh) * 2021-12-30 2022-01-28 青岛美迪康数字工程有限公司 Ct影像中肢体部位识别模型训练方法及系统
CN114612824A (zh) * 2022-03-09 2022-06-10 清华大学 目标识别方法及装置、电子设备和存储介质
CN116077066A (zh) * 2023-02-10 2023-05-09 北京安芯测科技有限公司 心电信号分类模型的训练方法、装置及电子设备

Also Published As

Publication number Publication date
KR20220009451A (ko) 2022-01-24
CN111368923B (zh) 2023-12-19
TWI770754B (zh) 2022-07-11
CN111368923A (zh) 2020-07-03
JP2022537974A (ja) 2022-08-31
TW202133787A (zh) 2021-09-16

Similar Documents

Publication Publication Date Title
WO2021174739A1 (zh) 神经网络训练方法及装置、电子设备和存储介质
WO2022151755A1 (zh) 目标检测方法及装置、电子设备、存储介质、计算机程序产品和计算机程序
WO2021147257A1 (zh) 网络训练、图像处理方法及装置、电子设备和存储介质
CN112767329B (zh) 图像处理方法及装置、电子设备
WO2021051965A1 (zh) 图像处理方法及装置、电子设备、存储介质和计算机程序
CN109886243B (zh) 图像处理方法、装置、存储介质、设备以及系统
WO2020211284A1 (zh) 图像处理方法及装置、电子设备和存储介质
WO2020211293A1 (zh) 一种图像分割方法及装置、电子设备和存储介质
JP2022537866A (ja) 画像分類方法、画像分類装置、画像処理方法、医療用電子機器、画像分類機器、及びコンピュータプログラム
Marostica et al. Development of a histopathology informatics pipeline for classification and prediction of clinical outcomes in subtypes of renal cell carcinoma
CN113222038B (zh) 基于核磁图像的乳腺病灶分类和定位方法及装置
US12118739B2 (en) Medical image processing method, apparatus, and device, medium, and endoscope
CN114820584B (zh) 肺部病灶定位装置
WO2023050691A1 (zh) 图像处理方法及装置、电子设备、存储介质和程序
CN112508918A (zh) 图像处理方法及装置、电子设备和存储介质
WO2021259390A2 (zh) 一种冠脉钙化斑块检测方法及装置
WO2023142532A1 (zh) 一种推理模型训练方法及装置
TW202346826A (zh) 影像處理方法
CN113902730A (zh) 图像处理和神经网络训练方法及装置
CN115170464A (zh) 肺图像的处理方法、装置、电子设备和存储介质
Wang et al. Breast cancer pre-clinical screening using infrared thermography and artificial intelligence: a prospective, multicentre, diagnostic accuracy cohort study
Saptasagar et al. Diagnosis and Prediction of Lung Tumour Using Combined ML Techniques
Mikos et al. An android-based pattern recognition application for the characterization of epidermal melanoma
CN113487537A (zh) 乳腺癌超声高回声晕的信息处理方法、装置及存储介质
Kumar et al. Automating cancer diagnosis using advanced deep learning techniques for multi-cancer image classification

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20923436

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2021574781

Country of ref document: JP

Kind code of ref document: A

ENP Entry into the national phase

Ref document number: 20217041454

Country of ref document: KR

Kind code of ref document: A

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20923436

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 27.03.2023)

122 Ep: pct application non-entry in european phase

Ref document number: 20923436

Country of ref document: EP

Kind code of ref document: A1