WO2020107156A1 - 乳腺医学超声图像的自动化分类方法及装置 - Google Patents

乳腺医学超声图像的自动化分类方法及装置 Download PDF

Info

Publication number
WO2020107156A1
WO2020107156A1 PCT/CN2018/117460 CN2018117460W WO2020107156A1 WO 2020107156 A1 WO2020107156 A1 WO 2020107156A1 CN 2018117460 W CN2018117460 W CN 2018117460W WO 2020107156 A1 WO2020107156 A1 WO 2020107156A1
Authority
WO
WIPO (PCT)
Prior art keywords
medical ultrasound
breast
target
ultrasound image
breast medical
Prior art date
Legal status (The legal status 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 status listed.)
Ceased
Application number
PCT/CN2018/117460
Other languages
English (en)
French (fr)
Inventor
王珊珊
肖韬辉
郑海荣
徐井旭
刘新
梁栋
李程
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Advanced Technology of CAS
Original Assignee
Shenzhen Institute of Advanced Technology of CAS
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 Shenzhen Institute of Advanced Technology of CAS filed Critical Shenzhen Institute of Advanced Technology of CAS
Priority to PCT/CN2018/117460 priority Critical patent/WO2020107156A1/zh
Publication of WO2020107156A1 publication Critical patent/WO2020107156A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition

Definitions

  • the present application relates to the technical field of image processing, and in particular to an automatic classification method and device for breast medical ultrasound images.
  • the way to classify breast medical images is to use a restricted Boltzmann machine (RBM) to design a two-layer feature extractor to extract and learn task-related features, and then use a support vector machine (SVM) to output the breast Image data feature type.
  • RBM restricted Boltzmann machine
  • SVM support vector machine
  • the two-layer feature extractor is not a deep learning model in the strict sense, the existing classification methods of breast medical images are not intelligent enough, and their shallow network results also lead to the quantity and quality of the extracted features All of them are limited. Therefore, the existing classification methods of breast medical images have the problems of low classification accuracy and poor performance.
  • the present application provides an automatic classification method and device for breast medical ultrasound images, which can effectively improve the accuracy and efficiency of feature recognition of breast medical ultrasound images, and can effectively and reliably improve breast medical ultrasound images.
  • the efficiency and intelligence of the image classification process and the accuracy of the classification results of breast medical ultrasound images are particularly important.
  • this application provides an automated classification method for breast medical ultrasound images, including:
  • HSV color space conversion is performed on the pre-processed target breast medical ultrasound image to obtain the hue H channel feature corresponding to the target breast medical ultrasound image;
  • the preprocessing of the acquired target breast medical ultrasound image includes:
  • the performing HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain the hue H-channel feature corresponding to the target breast medical ultrasound image includes:
  • HSV color space conversion is performed on the pre-processed target breast medical ultrasound image to obtain the HSV color feature corresponding to the target breast medical ultrasound image;
  • the automatic classification method of the breast medical ultrasound image further includes:
  • the deep convolutional neural network is composed of multiple convolutional layers and multiple fully connected layers in sequence. After the breast medical ultrasound image is input through the first convolutional layer, the deep convolutional neural network performs layer-by-layer Map to get different representations of the medical ultrasound images of the breast.
  • the acquiring the training sample set of the deep convolutional neural network based on multiple historical breast medical ultrasound images includes:
  • the h-channel values corresponding to each historical breast medical ultrasound image constitute the training sample set of the deep convolutional neural network.
  • the automatic classification method of breast medical ultrasound images further includes:
  • test sample based on at least one test breast medical ultrasound image
  • the current deep convolutional neural network is used as the target deep convolutional neural network for classifying breast medical ultrasound images.
  • the automatic classification method of breast medical ultrasound images further includes:
  • the current deep convolutional neural network does not meet the preset requirements, the current deep convolutional neural network is optimized and/or the updated training sample set is applied to re-train the deep convolutional neural network for model training.
  • this application provides an automatic classification device for breast medical ultrasound images, including:
  • the data preprocessing module is used to preprocess the acquired target breast medical ultrasound image
  • the channel feature extraction module is used to perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain the hue H channel feature corresponding to the target breast medical ultrasound image;
  • the model prediction module is used to use the hue H channel feature corresponding to the pre-processed target breast medical ultrasound image as a prediction sample, input a preset target deep convolutional neural network, and output the target deep convolutional neural network As the classification result of the target breast medical ultrasound image.
  • the data preprocessing module includes:
  • Target data acquisition unit used to acquire target breast medical ultrasound images
  • a target data cleaning unit used for performing data cleaning on the target breast medical ultrasound image
  • the target data labeling unit is used to mark the target area in the target breast medical ultrasound image after data cleaning to obtain a sample label of the target breast medical ultrasound image.
  • the channel feature extraction module includes:
  • a target HSV conversion unit configured to perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain the HSV color feature corresponding to the target breast medical ultrasound image;
  • the target H channel extraction unit is used to extract the hue H channel value corresponding to the target breast medical ultrasound image from the HSV color feature.
  • the automatic classification device for breast medical ultrasound images further includes:
  • the model building module is used to construct a deep convolutional neural network, and obtain the training sample set of the deep convolutional neural network based on multiple historical breast medical ultrasound images;
  • the model training module is configured to apply the training sample set to perform model training on the deep convolutional neural network.
  • the deep convolutional neural network is composed of multiple convolutional layers and multiple fully connected layers in sequence. After the breast medical ultrasound image is input through the first convolutional layer, the deep convolutional neural network is layered Map to get different representations of the medical ultrasound images of the breast.
  • model building module includes:
  • Historical data acquisition unit for acquiring multiple historical breast medical ultrasound images
  • Historical data cleaning unit used for data cleaning of all historical breast medical ultrasound images
  • the historical data labeling unit is used to mark the target areas in each historical breast medical ultrasound image after data cleaning, to obtain sample labels of each historical breast medical ultrasound image;
  • the historical HSV conversion unit is used to perform HSV color space conversion on each historical breast medical ultrasound image to obtain the HSV color features corresponding to each historical breast medical ultrasound image;
  • the historical H channel extraction unit is used to extract the hue H channel value corresponding to each historical breast medical ultrasound image from the HSV color features of each historical breast medical ultrasound image;
  • the training sample set generating unit is configured to compose the hue H channel value corresponding to each historical breast medical ultrasound image into the training sample set of the deep convolutional neural network.
  • the automatic classification device for breast medical ultrasound images further includes:
  • a test sample acquisition unit configured to obtain a test sample based on at least one breast medical ultrasound image for testing
  • a model testing unit used to apply the test samples to perform model testing on the deep convolutional neural network, and use the output of the deep convolutional neural network as a test result;
  • a test result judging unit for judging whether the current deep convolutional neural network meets the preset requirements based on the test result and the known classification result of at least one test breast medical ultrasound image, and if so, the current deep convolutional neural network As a target deep convolutional neural network for classifying breast medical ultrasound images.
  • test result judging unit is further used to optimize the current deep convolutional neural network and/or apply the updated training sample set if the current deep convolutional neural network does not meet the preset requirements. Deep convolutional neural network for model training.
  • the present application provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor.
  • the processor executes the program, it is implemented as follows:
  • HSV color space conversion is performed on the pre-processed target breast medical ultrasound image to obtain the hue H channel feature corresponding to the target breast medical ultrasound image;
  • the present application provides a computer-readable storage medium on which a computer program is stored, which is implemented when executed by a processor:
  • HSV color space conversion is performed on the pre-processed target breast medical ultrasound image to obtain the hue H channel feature corresponding to the target breast medical ultrasound image;
  • this application provides an automatic classification method and device for breast medical ultrasound images.
  • preprocessing the acquired target breast medical ultrasound images it can provide an accurate and reliable data foundation for subsequent image recognition.
  • the target breast medical ultrasound image is converted into HSV color space to obtain the hue H-channel feature corresponding to the target breast medical ultrasound image, which can extract high-throughput and high-level image features for breast ultrasound image feature classification.
  • the classification results of ultrasound images can effectively improve the accuracy and efficiency of the feature recognition of breast medical ultrasound images, and can effectively and reliably improve the efficiency and intelligence of the classification process of breast medical ultrasound images, and improve the breast medical ultrasound images.
  • the accuracy of the classification results can meet the automation needs of modern breast medicine and can effectively adapt to the current development trend of intelligent diagnosis of medical imaging big data.
  • FIG. 1 is a schematic structural diagram between a server S1 and a client device B1 in an embodiment of the present invention.
  • FIG. 2 is a schematic structural diagram of a server S1, a client device B1, and a database server S2 in an embodiment of the present invention.
  • FIG. 3 is a schematic flowchart of an automatic classification method of breast medical ultrasound images in an embodiment of the present invention.
  • FIG. 4 is a schematic flowchart of step 100 in an automated classification method of breast medical ultrasound images in an embodiment of the present invention.
  • FIG. 5 is a schematic flowchart of step 200 in an automatic classification method of breast medical ultrasound images in an embodiment of the present invention.
  • FIG. 6 is a schematic flowchart of an automatic classification method of breast medical ultrasound images including step 001 and step 002 in an embodiment of the present invention.
  • step 001 is a schematic flowchart of step 001 in an automatic classification method of breast medical ultrasound images in an embodiment of the present invention.
  • FIG. 8 is a schematic structural diagram of a deep convolutional neural network in an embodiment of the present invention.
  • FIG. 9 is a schematic flowchart of steps A01 to A05 in an automatic classification method of breast medical ultrasound images in an embodiment of the present invention.
  • FIG. 10 is a schematic flowchart of an apparatus for automatically classifying breast medical ultrasound images in an embodiment of the present invention.
  • FIG. 11 is a schematic flowchart of an automatic classification device for breast medical ultrasound images including a model building module 01 and a model training module 02 in an embodiment of the present invention.
  • FIG. 12 is a schematic flowchart of an automatic classification device for breast medical ultrasound images including a model test module A0 in an embodiment of the present invention.
  • FIG. 13 is a schematic structural diagram of an electronic device in an embodiment of the present invention.
  • the traditional method of classification of breast medical ultrasound images firstly needs to manually segment the region of interest of the breast image, and requires a wealth of prior knowledge to design features that can characterize different regions of interest, then through artificial feature extraction and selection, and finally through design Different classifiers distinguish regions of interest.
  • the process requires a lot of manual operations, and requires a priori knowledge.
  • the number of extracted features is extremely limited, and it is difficult to adapt to the current development trend of intelligent diagnosis of medical imaging big data.
  • an existing classification method of breast medical ultrasound images utilizes a restricted Boltzmann machine (RBM) to design a two-layer feature extractor to extract and learn task-related features, and finally uses support vectors
  • RBM restricted Boltzmann machine
  • SVM The machine outputs the feature type of breast image data.
  • This method has some limitations compared with the method proposed in this application.
  • this technique is not strictly a deep learning model, it only has a two-layer structure.
  • Second, its shallow network structure also leads to a gap in the quantity and quality of its extracted features from the technical solution of the present application, resulting in a large difference in the performance of feature classification. That is to say, the existing technology is mostly a traditional feature engineering method and a shallower network structure, and the extracted quantitative features are limited in number and level, which is the biggest disadvantage of the existing technology.
  • the accuracy of breast ultrasound image data classification is reduced, so improving the quality of its feature extraction is the starting point of the proposed technology.
  • the existing technology requires local segmentation of the breast image, which adds a lot of work; and the segmentation task itself It is not easy to complete, the segmentation result is directly the classification behind the image.
  • the feature extraction method also requires manual design, extraction and selection, which greatly increase the workload of classification.
  • this application provides an automatic classification method for breast medical ultrasound images, an automatic classification device for breast medical ultrasound images, electronic equipment, and a computer
  • the readable storage medium can provide an accurate and reliable data basis for subsequent image recognition by preprocessing the acquired target breast medical ultrasound image, and by performing HSV color space conversion on the preprocessed target breast medical ultrasound image, Obtain the tone H channel feature corresponding to the target breast medical ultrasound image, which can extract high-throughput, high-level image features for breast ultrasound image feature classification, by preprocessing the tone H channel corresponding to the target breast medical ultrasound image
  • the feature is used as a prediction sample, input a preset target deep convolutional neural network, and use the output of the target deep convolutional neural network as the classification result of the target breast medical ultrasound image, which can effectively improve the feature recognition of the breast medical ultrasound image Accuracy and efficiency, and can effectively and reliably improve the efficiency and intelligence of the classification process of breast medical ultrasound images, and improve the accuracy of the classification results of breast medical
  • the present application also provides an automatic classification device for breast medical ultrasound images.
  • the device may be a server S1.
  • the server S1 may be in communication with at least one client device B1.
  • the client The device B1 may send the target breast medical ultrasound image to the server S1, and the server S1 may receive the target breast medical ultrasound image online.
  • the server S1 may preprocess the acquired target breast medical ultrasound image online or offline, and perform HSV color space conversion on the preprocessed target breast medical ultrasound image to obtain the hue H channel corresponding to the target breast medical ultrasound image Feature, the tonal H-channel feature corresponding to the pre-processed target breast medical ultrasound image is used as a prediction sample, a preset target deep convolutional neural network is input, and the output of the target deep convolutional neural network is used as the target Breast medical ultrasound image classification results. Then, the server S1 may send the classification result of the target breast medical ultrasound image online to the client device B1. The client device B1 may receive the classification result of the target breast medical ultrasound image online.
  • the server S1 may also be in communication connection with at least one database server S2, where the database server S2 is used to store historical breast medical ultrasound image data.
  • the database server S2 online sends the historical breast medical ultrasound image to the server S1, the server S1 can receive the historical breast medical ultrasound image online, and then obtain the depth convolution according to a plurality of historical breast medical ultrasound images A training sample set of a neural network, and applying the training sample set to perform model training on the deep convolutional neural network.
  • the database server S2 may also be used to store breast medical ultrasound image data for testing.
  • the database server S2 online sends the test breast medical ultrasound image data to the server S1, the server S1 can receive the test breast medical ultrasound image data online, and then according to at least one test breast medical ultrasound image Obtain a test sample, and apply the test sample to perform a model test on the deep convolutional neural network, and use the output of the deep convolutional neural network as a test result, and then based on the test result and at least one test breast medical ultrasound
  • the known classification results of the image to determine whether the current deep convolutional neural network meets the preset requirements.
  • the current deep convolutional neural network is optimized and/or the updated training sample set is applied to perform model training on the deep convolutional neural network again.
  • the client device B1 may have a display interface so that the user can view the classification result of the target breast medical ultrasound image sent by the server S1 according to the interface.
  • the client device B1 may include a smart phone, a tablet electronic device, a network set-top box, a portable computer, a desktop computer, a personal digital assistant (PDA), a vehicle-mounted device, a smart wearable device, and the like.
  • the smart wearable device may include smart glasses, smart watches, smart bracelets and the like.
  • the automatic classification of breast medical ultrasound images can be performed on the server S1 side as described above, that is, the architecture as shown in FIG. 1, or all operations can be performed on the client device This is done in B1, and the client device B1 can directly communicate with the database server S2. Specifically, the selection may be based on the processing capability of the client device B1, and the user's usage scenario restrictions. This application does not limit this. If all operations are completed in the client device B1, the client device B1 may further include a processor for performing specific processing of automatic classification of breast medical ultrasound images.
  • the network protocol may include, for example, TCP/IP protocol, UDP/IP protocol, HTTP protocol, HTTPS protocol, and so on.
  • the network protocol may also include, for example, the RPC protocol (Remote Procedure Call Protocol), the REST protocol (Representational State Transfer, Representational State Transfer Protocol) used on top of the above protocols.
  • the test breast medical ultrasound image is not included in the historical breast medical ultrasound image used for model training, and for the test breast medical ultrasound image, it is necessary to obtain Its known classification results.
  • the breast medical ultrasound image may be a breast elastic ultrasound image, specifically embodied as breast elastic ultrasound image data.
  • This application uses an imaging omics method based on deep convolutional neural networks, which integrates the technical advantages of deep learning in image processing and the advantages of cutting-edge thinking in imaging omics.
  • Combining medical imaging big data, the medical ultrasound imaging data of breast Performing feature learning and extraction improves the ability to classify features of breast ultrasound image data.
  • the following embodiments and application scenarios are used for specific description.
  • the automatic classification method of breast medical ultrasound images specifically includes the following:
  • Step 100 Pre-process the acquired target breast medical ultrasound image.
  • Step 200 Perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain a hue H-channel feature corresponding to the target breast medical ultrasound image.
  • Step 300 Use the tone H-channel feature corresponding to the pre-processed target breast medical ultrasound image as a prediction sample, input a preset target deep convolutional neural network, and use the output of the target deep convolutional neural network as the Classification results of medical ultrasound images of target breast.
  • the target deep convolutional neural network is composed of multiple convolutional layers and multiple fully connected layers in sequence.
  • the deep convolutional neural network DCNN (Deep Convolutional Neural Network) is the first convolution of the breast medical ultrasound image. After layer input, layer-by-layer mapping is performed in the deep convolutional neural network to obtain different representations of the layers for the breast medical ultrasound image.
  • the last layers of the target deep convolutional neural network are composed of fully connected layers, and the activation function of the last layer is a softmax function, which is used to classify local features of breast ultrasound images.
  • the automatic classification method for breast medical ultrasound images extracts high-throughput and high-level image features through deep convolutional neural networks for feature classification of breast ultrasound images.
  • This application is based on the large sample data of breast ultrasound imaging, and uses the imaging omics method to screen out the key imaging histological features that characterize the clinical phenotype, and constructs a multi-parameter classification device for regions of interest (ROI) of breast ultrasound imaging.
  • ROI regions of interest
  • the present application also provides a specific implementation of step 100 in the automated classification method of breast medical ultrasound images, see Figure 4, the step 100 specifically includes the following:
  • Step 101 Acquire a target breast medical ultrasound image.
  • Step 102 Perform data cleaning on the target breast medical ultrasound image.
  • the automatic classification device for breast medical ultrasound images performs data cleaning on target breast elastic ultrasound image data, and deletes some system setting parameters and other information.
  • Step 103 Mark the target area in the target breast medical ultrasound image after data cleaning to obtain a sample label of the target breast medical ultrasound image.
  • the automatic classification device for medical ultrasound images of the breast marks the target area in the target mammary elastic ultrasound image data after data cleaning to obtain a sample label of the target mammary elastic ultrasound image data after data cleaning,
  • the target region may be a region of interest in breast elastic elasticity ultrasound image data of a doctor.
  • the present application also provides a specific implementation of step 200 in the automated classification method of breast medical ultrasound images, Referring to FIG. 5, the step 200 specifically includes the following content:
  • Step 201 Perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain the HSV color feature corresponding to the target breast medical ultrasound image.
  • Step 202 Extract the hue H channel value corresponding to the target breast medical ultrasound image from the HSV color feature.
  • this application provides a model prediction scenario of an automatic classification method for breast medical ultrasound images, the specific content is as follows:
  • the present application also provides model establishment and training performed before step 100 in the automatic classification method of breast medical ultrasound images For the process, see FIG. 6.
  • the model building and training process specifically includes the following:
  • Step 001 Construct a deep convolutional neural network, and obtain the training sample set of the deep convolutional neural network based on multiple historical breast medical ultrasound images.
  • Step 002 Apply the training sample set to perform model training on the deep convolutional neural network.
  • the step 001 of obtaining the training sample set of the deep convolutional neural network based on a plurality of historical breast medical ultrasound images specifically includes the following content:
  • Step 001a Acquire multiple historical breast medical ultrasound images
  • Step 002a Perform data cleaning on each historical breast medical ultrasound image
  • Step 003a Mark the target area in each historical breast medical ultrasound image after data cleaning to obtain a sample label of each historical breast medical ultrasound image
  • Step 004a Perform HSV color space conversion on each historical breast medical ultrasound image to obtain the HSV color features corresponding to each historical breast medical ultrasound image;
  • Step 005a extract, from the HSV color features of each historical breast medical ultrasound image, a hue H channel value corresponding to each historical breast medical ultrasound image;
  • Step 006a Combine the hue H channel values corresponding to each historical breast medical ultrasound image into the training sample set of the deep convolutional neural network.
  • this application provides a model establishment and training scenario of an automatic classification method for breast medical ultrasound images.
  • the specific content is as follows:
  • the deep convolutional neural network is composed of multiple convolutional layers and multiple fully connected layers: the input image is mapped layer by layer in the deep convolutional neural network to obtain different representations of the layers for the image to achieve the depth of the image.
  • the deep convolutional neural network specifically includes 5 convolutional layers: module 1 to module 5, and further specifically includes a module 6 composed of 3 fully connected layers, and the modules 1 to 6 are arranged in sequence.
  • the specific result of the deep convolutional neural network is shown in FIG. 8, and the specific content in each module is shown in Table 1 below.
  • * represents the convolution operation between the input of each layer and the filter; Is the output of the jth neuron after the convolution of the lth layer; Is the output of the k-th neuron in layer l-1, that is, the input data of layer l; Is the weight of the k-th neuron in layer l-1 connected to the j-th neuron in layer l; c is the offset corresponding to the j-th neuron in layer l;
  • there are ReLU functions of other transformation forms such as Leaky-ReLU, P-ReLU, etc.
  • the last layers of the deep convolutional neural network designed by this application are composed of fully connected layers.
  • the activation function of the last layer is the softmax function, which is used to classify the local features of breast ultrasound images.
  • W L and b L are the weight and offset of the last fully connected layer, respectively.
  • the process of acquiring training samples may be performed in parallel with the process of establishing a deep convolutional neural network, or may be performed before or after the process of establishing a deep convolutional neural network.
  • the hue H channel value corresponding to each historical breast elastic ultrasound image data constitutes the training sample set of the deep convolutional neural network.
  • Model training is performed on the deep convolutional neural network with a training sample set composed of hue H channel values corresponding to the historical breast elastic ultrasound image data respectively.
  • the present application also provides the automatic classification method of the breast medical ultrasound images performed before step 100 and after step 002
  • the model testing process specifically includes the following:
  • Step A01 Obtain a test sample based on at least one breast medical ultrasound image for testing
  • Step A02 apply the test sample to perform a model test on the deep convolutional neural network, and use the output of the deep convolutional neural network as the test result;
  • Step A03 based on the test result and the known classification result of at least one test breast medical ultrasound image, determine whether the current deep convolutional neural network meets the preset requirements;
  • step A04 If yes, go to step A04, otherwise, go to step A05.
  • Step A04 The current deep convolutional neural network is used as the target deep convolutional neural network for classifying breast medical ultrasound images.
  • Step A05 If the current deep convolutional neural network does not meet the preset requirements, optimize the current deep convolutional neural network and/or apply the updated training sample set to perform model training on the deep convolutional neural network again.
  • this application provides a model test scenario of an automated classification method for breast medical ultrasound images, the specific contents are as follows:
  • the Hue H channel value corresponding to the breast elastic ultrasound image data for each test is used as the test sample of the deep convolutional neural network.
  • the specific method for optimizing the deep convolutional neural network in the process of returning to the establishment of the deep convolutional neural network is as follows:
  • the weights and offsets are updated by BP back propagation algorithm, and the parameters of deep convolutional neural network are optimized by optimizing the loss function of classification cross entropy, as shown in equation (4), It can extract the features of the region of interest in the energetic breast ultrasound image, and achieve the purpose of rapid and accurate classification of the features of the region of interest in the breast ultrasound image.
  • I and K are the number of classification categories and the total number of training samples
  • Y′ i is the label of the healthy area or the lesion area.
  • the automatic classification method of breast medical ultrasound images provided by the embodiments of the present application can be used to classify the features of the breast region of interest, such as the characteristics of the healthy area or the lesion area or the suspicious lesion area, but it is not limited to the position of the breast .
  • the technology of this application helps to reduce the workload of clinicians and improve the efficiency of doctors by analyzing breast medical image data; in addition, the method can classify breast medical image features with its advantages of non-invasive, real-time, safe and convenient To apply.
  • an automatic classification device for breast medical ultrasound images for implementing the entire content of the automatic classification method for breast medical ultrasound images, see FIG. 10, the automatic classification device for breast medical ultrasound images specifically It includes the following:
  • the data pre-processing module 10 is used to pre-process the acquired target breast medical ultrasound image.
  • the channel feature extraction module 20 is configured to perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain a hue H-channel feature corresponding to the target breast medical ultrasound image.
  • the model prediction module 30 is configured to use the tone H-channel feature corresponding to the target breast medical ultrasound image as a prediction sample, input a preset target depth convolutional neural network, and convert the target depth convolutional neural network The classification result of the target breast medical ultrasound image is output.
  • the embodiment of the automatic classification apparatus for breast medical ultrasound images provided by the present application may be specifically used to execute all the processing processes of each embodiment of the automatic classification method for breast medical ultrasound images in the above embodiments, and the functions thereof will not be repeated here. Reference may be made to the detailed description of the above method embodiments.
  • the automatic classification device for breast medical ultrasound images preprocesses the acquired target breast medical ultrasound images through the data preprocessing module 10, which can provide an accurate and reliable data basis for subsequent image recognition .
  • the channel feature extraction module 20 performs HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain the hue H-channel feature corresponding to the target breast medical ultrasound image, which can extract high-throughput, high-level image features
  • the model prediction module 30 uses the hue H channel feature corresponding to the target breast medical ultrasound image as a prediction sample, enters a preset target depth convolutional neural network, and sets the target depth
  • the output of the convolutional neural network as the classification result of the target breast medical ultrasound image can effectively improve the accuracy and efficiency of the feature recognition of the breast medical ultrasound image, and can effectively and reliably improve the efficiency of the classification process of the breast medical ultrasound image
  • the degree of sexuality and intelligence, and improve the accuracy of the classification results of breast medical ultrasound images which can meet the automation
  • the present application also provides a specific implementation of the data preprocessing module 10 in the automatic classification device for breast medical ultrasound images Way, the data preprocessing module 10 specifically includes the following content:
  • the target data acquisition unit 11 is used to acquire a target breast medical ultrasound image.
  • the target data cleaning unit 12 is used for performing data cleaning on the target breast medical ultrasound image.
  • the target data labeling unit 13 is configured to mark the target area in the target breast medical ultrasound image after data cleaning to obtain a sample label of the target breast medical ultrasound image.
  • the present application also provides specifics of the channel feature extraction module 20 in the automatic classification device for breast medical ultrasound images
  • the channel feature extraction module 20 specifically includes the following:
  • the target HSV conversion unit 21 is configured to perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain the HSV color feature corresponding to the target breast medical ultrasound image.
  • the target H channel extraction unit 22 is configured to extract the hue H channel value corresponding to the target breast medical ultrasound image from the HSV color feature.
  • the present application also provides a model building module and a model training module in the automatic classification device for breast medical ultrasound images, see Figure 11, the model building module and model training module specifically include the following:
  • the model building module 01 is used to construct a deep convolutional neural network, and obtain the training sample set of the deep convolutional neural network according to multiple historical breast medical ultrasound images.
  • the model training module 02 is configured to apply the training sample set to perform model training on the deep convolutional neural network.
  • the deep convolutional neural network is composed of multiple convolutional layers and multiple fully connected layers in sequence. After the breast medical ultrasound image is input through the first convolutional layer, it is performed in the deep convolutional neural network Layer by layer mapping to obtain different representations of each layer for the breast medical ultrasound image.
  • the model building module 01 also includes the following contents:
  • the historical data acquisition unit 01a is used to acquire multiple historical breast medical ultrasound images
  • the historical data cleaning unit 01b is used to perform data cleaning on each historical breast medical ultrasound image
  • the historical data labeling unit 01c is used to mark the target areas in each historical breast medical ultrasound image after data cleaning to obtain sample labels of each historical breast medical ultrasound image;
  • the historical HSV conversion unit 01d is used to perform HSV color space conversion on each historical breast medical ultrasound image to obtain HSV color features corresponding to each historical breast medical ultrasound image;
  • the historical H channel extraction unit 01e is used to extract the hue H channel value corresponding to each historical breast medical ultrasound image from the HSV color features of each historical breast medical ultrasound image;
  • the training sample set generation unit 01f is configured to compose the hue H channel value corresponding to each historical breast medical ultrasound image into the training sample set of the deep convolutional neural network.
  • the present application also provides a model test module A0 in the automatic classification device for breast medical ultrasound images.
  • the model test module A0 specifically includes the following content:
  • test sample acquisition unit A1 which is used to obtain a test sample based on at least one breast medical ultrasound image for testing
  • a model testing unit A2 configured to apply the test samples to perform model testing on the deep convolutional neural network, and use the output of the deep convolutional neural network as a test result;
  • a test result determination unit A3 used to determine whether the current deep convolutional neural network meets the preset requirements based on the test result and the known classification results of at least one test breast medical ultrasound image, if so, the current deep convolutional neural network
  • the network serves as a target deep convolutional neural network for classifying breast medical ultrasound images.
  • the test result judgment unit is also used to optimize the current deep convolutional neural network and/or apply the updated training sample set to re-contract the depth if the current deep convolutional neural network does not meet the preset requirements.
  • the automatic classification device for breast medical ultrasound images can be used to classify the characteristics of the breast region of interest, such as the characteristics of the healthy area or the lesion area or the suspicious lesion area, but it is not limited to the position of the breast .
  • the purpose of automatic classification of breast image features is achieved.
  • the technology of this application helps to reduce the workload of clinicians and improve the efficiency of doctors by analyzing breast medical image data; in addition, the device can classify breast medical image features with its advantages of non-invasive, real-time, safe and convenient To apply.
  • Embodiments of the present application also provide a specific implementation of an electronic device that can implement all steps in the automatic classification method of breast medical ultrasound images in the foregoing embodiments.
  • the electronic device specifically includes the following:
  • Processor processing 601, memory (memory) 602, communications interface (Communications) Interface 603 and bus 604;
  • the processor 601, the memory 602, and the communication interface 603 communicate with each other through the bus 604; the communication interface 603 is used to implement an automatic classification device, server, client terminal, and other participating institutions for breast medical ultrasound images Information transmission between;
  • the processor 601 is used to call a computer program in the memory 602, and when the processor executes the computer program, it implements all the steps in the automatic classification method for breast medical ultrasound images in the foregoing embodiments, for example, the When the processor executes the computer program, the following steps are realized:
  • Step 100 Pre-process the acquired target breast medical ultrasound image.
  • Step 200 Perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain a hue H-channel feature corresponding to the target breast medical ultrasound image.
  • Step 300 Use the tone H-channel feature corresponding to the pre-processed target breast medical ultrasound image as a prediction sample, input a preset target deep convolutional neural network, and use the output of the target deep convolutional neural network as the Classification results of medical ultrasound images of target breast.
  • the electronic device provided by the embodiment of the present application can be used to classify the characteristics of the breast region of interest, such as the characteristics of the healthy region or the diseased region or the suspicious diseased region, but it is not limited to the position of the breast.
  • the electronic device can be used to classify the characteristics of the breast region of interest, such as the characteristics of the healthy region or the diseased region or the suspicious diseased region, but it is not limited to the position of the breast.
  • the electronic device provided by the embodiment of the present application can be used to classify the characteristics of the breast region of interest, such as the characteristics of the healthy region or the diseased region or the suspicious diseased region, but it is not limited to the position of the breast.
  • the technology of this application helps to reduce the workload of clinicians and improve the efficiency of doctors by analyzing breast medical image data; in addition, the device can classify breast medical image features with its advantages of non-invasive, real-time, safe and convenient To apply.
  • An embodiment of the present application also provides a computer-readable storage medium capable of implementing all the steps in the automated classification method for breast medical ultrasound images in the above embodiments
  • the computer-readable storage medium stores a computer program
  • the computer program When executed by the processor, all steps of the automatic classification method for breast medical ultrasound images in the foregoing embodiments are implemented. For example, when the processor executes the computer program, the following steps are realized:
  • Step 100 Pre-process the acquired target breast medical ultrasound image.
  • Step 200 Perform HSV color space conversion on the pre-processed target breast medical ultrasound image to obtain a hue H-channel feature corresponding to the target breast medical ultrasound image.
  • Step 300 Use the tone H-channel feature corresponding to the pre-processed target breast medical ultrasound image as a prediction sample, input a preset target deep convolutional neural network, and use the output of the target deep convolutional neural network as the Classification results of medical ultrasound images of target breast.
  • the computer-readable storage medium provided by the embodiments of the present application can be used to classify the features of the breast region of interest, such as the characteristics of the healthy region or the lesion region or the suspicious lesion region, but it is not limited to the location of the breast.
  • the features of the breast region of interest such as the characteristics of the healthy region or the lesion region or the suspicious lesion region, but it is not limited to the location of the breast.
  • the technology of this application helps to reduce the workload of clinicians and improve the efficiency of doctors by analyzing breast medical image data; in addition, the device can classify breast medical image features with its advantages of non-invasive, real-time, safe and convenient To apply.
  • the system, device, module or unit explained in the above embodiments may be specifically implemented by a computer chip or entity, or implemented by a product with a certain function.
  • a typical implementation device is a computer.
  • the computer may be, for example, a personal computer, a laptop computer, an on-board human-machine interaction device, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet A computer, a wearable device, or any combination of these devices.
  • the functions are divided into various modules and described separately.
  • the functions of each module may be implemented in one or more software and/or hardware, or the modules that implement the same function may be implemented by a combination of multiple submodules or subunits.
  • the device embodiments described above are only schematic.
  • the division of the unit is only a division of logical functions.
  • there may be another division manner for example, multiple units or components may be combined or integrated To another system, or some features can be ignored, or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
  • controller in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program method steps to make the controller use logic gates, switches, application specific integrated circuits, programmable logic controllers and embedded The same function is realized in the form of a microcontroller or the like. Therefore, such a controller can be regarded as a hardware component, and the device for implementing various functions included therein can also be regarded as a structure within the hardware component. Or even, the means for realizing various functions can be regarded as both a software module of an implementation method and a structure within a hardware component.
  • each flow and/or block in the flowchart and/or block diagram and a combination of the flow and/or block in the flowchart and/or block diagram may be implemented by computer program instructions.
  • These computer program instructions can be provided to the processor of a general-purpose computer, special-purpose computer, embedded processing machine, or other programmable data processing device to produce a machine that enables the generation of instructions executed by the processor of the computer or other programmable data processing device
  • These computer program instructions may also be stored in a computer-readable memory that can guide a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, the instructions
  • the device implements the functions specified in one block or multiple blocks of the flowchart one flow or multiple flows and/or block diagrams.
  • These computer program instructions can also be loaded onto a computer or other programmable data processing device, so that a series of operating steps are performed on the computer or other programmable device to produce computer-implemented processing, which is executed on the computer or other programmable device
  • the instructions provide steps for implementing the functions specified in one block or multiple blocks of the flowchart one flow or multiple flows and/or block diagrams.
  • the computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.
  • processors CPUs
  • input/output interfaces network interfaces
  • memory volatile and non-volatile memory
  • the memory may include non-permanent memory, random access memory (RAM) and/or non-volatile memory in computer-readable media, such as read only memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.
  • RAM random access memory
  • ROM read only memory
  • flash RAM flash random access memory
  • Computer-readable media including permanent and non-permanent, removable and non-removable media, can store information by any method or technology.
  • the information may be computer readable instructions, data structures, modules of programs, or other data.
  • Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, read-only compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, Magnetic tape cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media can be used to store information that can be accessed by computing devices.
  • computer-readable media does not include temporary computer-readable media (transitory media), such as modulated data signals and carrier waves.
  • the embodiments of the present specification may be provided as methods, systems, or computer program products. Therefore, the embodiments of the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Moreover, the embodiments of the present specification may take the form of computer program products implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.
  • computer usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
  • Embodiments of this specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules.
  • program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types.
  • the embodiments of the present specification may also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected through a communication network.
  • program modules may be located in local and remote computer storage media including storage devices.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Ultra Sonic Daignosis Equipment (AREA)

Abstract

一种乳腺医学超声图像的自动化分类方法及装置，方法包括：对获取的目标乳腺医学超声图像进行预处理（000）；对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征（100）；将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果（200）。该方法能够有效提高对乳腺医学超声图像的特征识别的准确性及效率，并能够有效可靠提高对乳腺医学超声图像的分类过程的高效性及智能化程度，以及提高乳腺医学超声图像的分类结果的准确性。

Description

乳腺医学超声图像的自动化分类方法及装置 技术领域
本申请涉及充图像处理技术领域，具体涉及一种乳腺医学超声图像的自动化分类方法及装置。
背景技术
随着科技在医学领域的飞速发展，通过人工对乳腺医学图像进行区域分割、特征提取及分类的技术已无法满足现代化乳腺医学的自动化需求。因此，为了有效适应现在医学影像大数据智能诊断的发展趋势，如何对乳腺医学图像进行智能化程度更高的分类已成为了重要的医学课题。
现有技术中，对乳腺医学图像进行分类的方式为利用限制玻尔兹曼机(RBM)设计个双层特征提取器提取和学习与任务相关的特征，而后利用支持向量机(SVM)输出乳腺影像数据特征类型。
然而，由于双层特征提取器并非严格意义上的深度学习模型，使得现有的乳腺医学图像的分类方法的智能化程度不足，且其较浅的网络结果也导致其提取特征在数量和质量上均受限，因此，现有的乳腺医学图像的分类方法存在分类准确性低及性能差的问题。
发明内容
针对现有技术中的问题，本申请提供一种乳腺医学超声图像的自动化分类方法及装置，能够有效提高对乳腺医学超声图像的特征识别的准确性及效率，并能够有效可靠提高对乳腺医学超声图像的分类过程的高效性及智能化程度，以及提高乳腺医学超声图像的分类结果的准确性。
为解决上述技术问题，本申请提供以下技术方案：
第一方面，本申请提供一种乳腺医学超声图像的自动化分类方法，包括：
对获取的目标乳腺医学超声图像进行预处理；
对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征；
将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
进一步地，所述对获取的目标乳腺医学超声图像进行预处理，包括：
获取目标乳腺医学超声图像；
对所述目标乳腺医学超声图像进行数据清洗；
对经数据清洗后的目标乳腺医学超声图像中的目标区域进行标记，得到所述目标乳腺医学超声图像的样本标签。
进一步地，所述对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征，包括：
对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的HSV颜色特征；
在所述HSV颜色特征中提取所述目标乳腺医学超声图像对应的色调H通道值。
进一步地，在所述对获取的目标乳腺医学超声图像进行预处理之前，所述乳腺医学超声图像的自动化分类方法还包括：
构建一深度卷积神经网络，以及，根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集；
应用所述训练样本集对所述深度卷积神经网络进行模型训练。
进一步地，所述深度卷积神经网络由多个卷积层和多个全连接层依次组成，乳腺医学超声图像经首个卷积层输入后，在所述深度卷积神经网络中进行层层映射，得到各层对于该乳腺医学超声图像不同的表示形式。
进一步地，所述根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集，包括：
获取多个历史乳腺医学超声图像；
对各个历史乳腺医学超声图像均进行数据清洗；
对经数据清洗的各个历史乳腺医学超声图像中的目标区域分别进行标记，得到各个历史乳腺医学超声图像的样本标签；
将各个历史乳腺医学超声图像进行HSV颜色空间转换，得到各个历史乳腺医学超声图像分别对应的HSV颜色特征；
在各个历史乳腺医学超声图像的所述HSV颜色特征中，分别提取各个历史乳腺医学超声图像对应的色调H通道值；
将各个历史乳腺医学超声图像对应的色调H通道值组成所述深度卷积神经网络的训练样本集。
进一步地，所述乳腺医学超声图像的自动化分类方法还包括：
根据至少一个测试用乳腺医学超声图像得到测试样本；
应用所述测试样本对所述深度卷积神经网络进行模型测试，并将该深度卷积神经网络的输出作为测试结果；
基于所述测试结果及至少一个测试用乳腺医学超声图像的已知分类结果，判断当前深度卷积神经网络是否符合预设要求；
若是，则将当前深度卷积神经网络作为用于对乳腺医学超声图像进行分类的目标深度卷积神经网络。
进一步地，所述乳腺医学超声图像的自动化分类方法还包括：
若当前深度卷积神经网络不符合所述预设要求，则对当前深度卷积神经网络进行优化和/或应用更新后的训练样本集重新对该深度卷积神经网络进行模型训练。
第二方面，本申请提供一种乳腺医学超声图像的自动化分类装置，包括：
数据预处理模块，用于对获取的目标乳腺医学超声图像进行预处理；
通道特征提取模块，用于对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征；
模型预测模块，用于将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
进一步地，所述数据预处理模块，包括：
目标数据获取单元，用于获取目标乳腺医学超声图像；
目标数据清洗单元，用于对所述目标乳腺医学超声图像进行数据清洗；
目标数据标记单元，用于对经数据清洗后的目标乳腺医学超声图像中的目标区域进行标记，得到所述目标乳腺医学超声图像的样本标签。
进一步地，所述通道特征提取模块包括：
目标HSV转换单元，用于对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的HSV颜色特征；
目标H通道提取单元，用于在所述HSV颜色特征中提取所述目标乳腺医学超声图像对应的色调H通道值。
进一步地，所述乳腺医学超声图像的自动化分类装置还包括：
模型构建模块，用于构建一深度卷积神经网络，以及，根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集；
模型训练模块，用于应用所述训练样本集对所述深度卷积神经网络进行模型训练。
进一步地，所述深度卷积神经网络由多个卷积层和多个全连接层依次组成，乳腺医学超声图像经首个卷积层输入后，在所述深度卷积神经网络中进行层层映射，得到各层对于该乳腺医学超声图像不同的表示形式。
进一步地，所述模型构建模块包括：
历史数据获取单元，用于获取多个历史乳腺医学超声图像；
历史数据清洗单元，用于对各个历史乳腺医学超声图像均进行数据清洗；
历史数据标记单元，用于对经数据清洗的各个历史乳腺医学超声图像中的目标区域分别进行标记，得到各个历史乳腺医学超声图像的样本标签；
历史HSV转换单元，用于将各个历史乳腺医学超声图像进行HSV颜色空间转换，得到各个历史乳腺医学超声图像分别对应的HSV颜色特征；
历史H通道提取单元，用于在各个历史乳腺医学超声图像的所述HSV颜色特征中，分别提取各个历史乳腺医学超声图像对应的色调H通道值；
训练样本集生成单元，用于将各个历史乳腺医学超声图像对应的色调H通道值组成所述深度卷积神经网络的训练样本集。
进一步地，所述乳腺医学超声图像的自动化分类装置还包括：
测试样本获取单元，用于根据至少一个测试用乳腺医学超声图像得到测试样本；
模型测试单元，用于应用所述测试样本对所述深度卷积神经网络进行模型测试，并将该深度卷积神经网络的输出作为测试结果；
测试结果判定单元，用于基于所述测试结果及至少一个测试用乳腺医学超声图像的已知分类结果，判断当前深度卷积神经网络是否符合预设要求，若是，则将当前深度卷积神经网络作为用于对乳腺医学超声图像进行分类的目标深度卷积神经网络。
进一步地，所述测试结果判定单元还用于若当前深度卷积神经网络不符合所述预设要求，则对当前深度卷积神经网络进行优化和/或应用更新后的训练样本集重新对该深度卷积神经网络进行模型训练。
第三方面，本申请提供一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现：
对获取的目标乳腺医学超声图像进行预处理；
对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征；
将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
第四方面，本申请提供一种计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现：
对获取的目标乳腺医学超声图像进行预处理；
对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征；
将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
由上述技术方案可知，本申请提供一种乳腺医学超声图像的自动化分类方法及装置，通过对获取的目标乳腺医学超声图像进行预处理，能够为后续图像识别提供准确且可靠的数据基础，通过对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征，能够提取高通量、高层次影像特征用于乳腺超声图像特征分类，通过将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果，能够有效提高对乳腺医学超声图像的特征识别的准确性及效率，并能够有效可靠提高对乳腺医学超声图像的分类过程的高效性及智能化程度，以及提高乳腺医学超声图像的分类结果的准确性，进而能够满足现代化乳腺医学的自动化需求并能够有效适应现在医学影像大数据智能诊断的发展趋势。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例中的服务器S1与客户端设备B1之间的构架示意图。
图2为本发明实施例中的服务器S1、客户端设备B1及数据库服务器S2之间的构架示意图。
图3为本发明实施例中的乳腺医学超声图像的自动化分类方法的流程示意图。
图4为本发明实施例中的乳腺医学超声图像的自动化分类方法中步骤100的流程示意图。
图5为本发明实施例中的乳腺医学超声图像的自动化分类方法中步骤200的流程示意图。
图6为本发明实施例中的包含有步骤001和步骤002的乳腺医学超声图像的自动化分类方法的流程示意图。
图7为本发明实施例中的乳腺医学超声图像的自动化分类方法中步骤001的流程示意图。
图8为本发明实施例中的深度卷积神经网络的结构示意图。
图9为本发明实施例中的乳腺医学超声图像的自动化分类方法中步骤A01至步骤A05的流程示意图。
图10为本发明实施例中的乳腺医学超声图像的自动化分类装置的流程示意图。
图11为本发明实施例中的包含有模型构建模块01和模型训练模块02的乳腺医学超声图像的自动化分类装置的流程示意图。
图12为本发明实施例中的包含有模型测试模块A0的乳腺医学超声图像的自动化分类装置的流程示意图。
图13为本发明实施例中的电子设备的结构示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚明白，下面结合附图对本发明实施例做进一步详细说明。在此，本发明的示意性实施例及其说明用于解释本发明，但并不作为对本发明的限定。
传统的乳腺医学超声图像的分类方法首先需要人工对乳腺图像感兴趣区域进行分割，并且需要丰富的先验知识来设计可以表征不同感兴趣区域的特征，再通过人工特征提取和选择，最后通过设计不同的分类器对感兴趣区域进行区分。已有技术中，其过程需要很多手工操作，而且需要先验知识，同时其提取的特征数量也极为有限，难以适合现在医学影像大数据智能诊断的发展趋势。而在此基础上，现有的一种乳腺医学超声图像的分类方法利用限制玻尔兹曼机(RBM)，设计了一个双层特征提取器提取和学习与任务相关的特征，最后利用支持向量机(SVM)输出乳腺影像数据特征类型。该方法与本申请所提出方法相比存在一些局限，首先此技术不是严格意义上的深度学习模型，它只有两层的结构。第二，其较浅的网络结构也导致其提取特征在数量和质量上与本申请技术方案的差距，导致特征分类的性能差异较大。也就是说，现有技术多为传统特征工程的方法和较浅层次的网络结构，其提取的量化特征数量有限、等级有限，这是现有技术的最大缺点。从而导致了乳腺超声影像数据分类的准确率下降，所以改进其特征提取的质量是所提出技术的出发点。此外，现有技术对数据预处理和特征提取方式上，存在许多不必要的工作，首先，在数据预处理方面现有技术需要对乳腺图像进行局部分割，增加了很多工作量；而且分割任务本身不容易完成，分割结果直接影像后面的分类。其次，特征提取方式也需要手工设计、提取及选择，这些都大大增加了分类的工作量。
考虑到上述现有的乳腺医学图像的分类方法存在分类准确性低及性能差的问题，本申请提供一种乳腺医学超声图像的自动化分类方法、乳腺医学超声图像的自动化分类装置、电子设备及计算机可读存储介质，通过对获取的目标乳腺医学超声图像进行预处理，能够为后续图像识别提供准确且可靠的数据基础，通过对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征，能够提取高通量、高层次影像特征用于乳腺超声图像特征分类，通过将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果，能够有效提高对乳腺医学超声图像的特征识别的准确性及效率，并能够有效可靠提高对乳腺医学超声图像的分类过程的高效性及智能化程度，以及提高乳腺医学超声图像的分类结果的准确性，进而能够满足现代化乳腺医学的自动化需求并能够有效适应现在医学影像大数据智能诊断的发展趋势。
基于上述内容，本申请还提供一种乳腺医学超声图像的自动化分类装置，该装置可以为一种服务器S1，参见图1，该服务器S1可以与至少一个客户端设备B1通信连接， 所述客户端设备B1可以将目标乳腺医学超声图像发送至所述服务器S1，所述服务器S1可以在线接收所述目标乳腺医学超声图像。所述服务器S1可以在线或者离线对获取的目标乳腺医学超声图像进行预处理，对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征，将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。而后，所述服务器S1可以将所述目标乳腺医学超声图像的分类结果在线发送至所述客户端设备B1。所述客户端设备B1可以在线接收所述目标乳腺医学超声图像的分类结果。
另外，参见图2，所述服务器S1还可以与至少一个数据库服务器S2通信连接，所述数据库服务器S2用于存储历史乳腺医学超声图像数据。所述数据库服务器S2在线将所述历史乳腺医学超声图像发送至所述服务器S1，所述服务器S1可以在线接收所述历史乳腺医学超声图像，而后根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集，应用所述训练样本集对所述深度卷积神经网络进行模型训练。
基于上述内容，所述数据库服务器S2还可以用于存储测试用乳腺医学超声图像数据。所述数据库服务器S2在线将所述测试用乳腺医学超声图像数据发送至所述服务器S1，所述服务器S1可以在线接收所述测试用乳腺医学超声图像数据，而后根据至少一个测试用乳腺医学超声图像得到测试样本，并应用所述测试样本对所述深度卷积神经网络进行模型测试，并将该深度卷积神经网络的输出作为测试结果，再基于所述测试结果及至少一个测试用乳腺医学超声图像的已知分类结果，判断当前深度卷积神经网络是否符合预设要求，若是，则将当前深度卷积神经网络作为用于对乳腺医学超声图像进行分类的目标深度卷积神经网络；若当前深度卷积神经网络不符合所述预设要求，则对当前深度卷积神经网络进行优化和/或应用更新后的训练样本集重新对该深度卷积神经网络进行模型训练。
基于上述内容，所述客户端设备B1可以具有显示界面，使得用户能够根据界面查看所述服务器S1发送的所述目标乳腺医学超声图像的分类结果。
可以理解的是，所述客户端设备B1可以包括智能手机、平板电子设备、网络机顶盒、便携式计算机、台式电脑、个人数字助理(PDA)、车载设备、智能穿戴设备等。其中，所述智能穿戴设备可以包括智能眼镜、智能手表、智能手环等。
在实际应用中，进行乳腺医学超声图像的自动化分类的部分可以在如上述内容所述的服务器S1侧执行，即，如图1所示的架构，也可以所有的操作都在所述客户端设备B1中完成，且该所述客户端设备B1可以直接与数据库服务器S2进行通信连接。具体可以根据所述客户端设备B1的处理能力，以及用户使用场景的限制等进行选择。本申请对此不作限定。若所有的操作都在所述客户端设备B1中完成，所述客户端设备B1还可以包括处理器，用于进行乳腺医学超声图像的自动化分类的具体处理。
所述服务器与所述客户端设备之间可以使用任何合适的网络协议进行通信，包括在本申请提交日尚未开发出的网络协议。所述网络协议例如可以包括TCP/IP协议、UDP/IP协议、HTTP协议、HTTPS协议等。当然，所述网络协议例如还可以包括在上述协议之上使用的RPC协议(Remote Procedure Call Protocol，远程过程调用协议)、REST协议(Representational State Transfer，表述性状态转移协议)等。
在本申请的一个或多个实施例中，所述测试用乳腺医学超声图像为未包含在用于模型训练的历史乳腺医学超声图像中的，且针对所述测试用乳腺医学超声图像，需获取其已知分类结果。
在本申请的一个或多个实施例中，所述乳腺医学超声图像可以为乳腺弹性超声图像，具体体现为乳腺弹性超声影像数据。
本申请采用基于深度卷积神经网络的影像组学方法，整合了深度学习在图像处方面的技术优势和影像组学前沿思路方面的优势，并结合医学影像大数据，通过对乳腺医学超声影像数据进行特征学习、提取，提高了对乳腺超声影像数据的特征分类能力。具体通过下述实施例及应用场景进行具体说明。
为了能够有效提高对乳腺医学超声图像的特征识别的准确性及效率，并能够有效可靠提高对乳腺医学超声图像的分类过程的高效性及智能化程度，以及提高乳腺医学超声图像的分类结果的准确性，本申请实施例提供一种乳腺医学超声图像的自动化分类方法，参见图3，所述乳腺医学超声图像的自动化分类方法具体包括如下内容：
步骤100：对获取的目标乳腺医学超声图像进行预处理。
步骤200：对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征。
步骤300：将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
可以理解的是，所述目标深度卷积神经网络由多个卷积层和多个全连接层依次组成的深度卷积神经网络DCNN(Deep convolutional neural network)，乳腺医学超声图像经首个卷积层输入后，在所述深度卷积神经网络中进行层层映射，得到各层对于该乳腺医学超声图像不同的表示形式。所述目标深度卷积神经网络的最后几层通过全连接层构成，最后一层的激活函数为softmax函数，用于乳腺超声图像局部特征的分类。
从上述描述可知，本申请实施例提供的乳腺医学超声图像的自动化分类方法，通过深度卷积神经网络提取高通量、高层次影像特征用于乳腺超声图像特征分类。本申请基于乳腺超声成像大样本数据，利用影像组学方法筛选出表征临床表型的关键影像组学特征，构建乳腺超声影像感兴趣区域ROI(region of interest)的多参数分类装置。
为了通过提高数据基础的准确性来进一步有效提高图像分类的准确性，在本申请的一实施例中，本申请还提供所述乳腺医学超声图像的自动化分类方法中步骤100的具体实施方式，参见图4，所述步骤100具体包括如下内容：
步骤101：获取目标乳腺医学超声图像。
步骤102：对所述目标乳腺医学超声图像进行数据清洗。
可以理解的是，所述乳腺医学超声图像的自动化分类装置对目标乳腺弹性超声影像数据进行数据清洗，删除一些系统设置参数等信息。
步骤103：对经数据清洗后的目标乳腺医学超声图像中的目标区域进行标记，得到所述目标乳腺医学超声图像的样本标签。
可以理解的是，所述乳腺医学超声图像的自动化分类装置对经数据清洗后的目标乳腺弹性超声影像数据中的目标区域进行标记，得到经数据清洗后的目标乳腺弹性超声影像数据的样本标签，其中，所述目标区域可以为医生对乳腺弹性超声影像数据中的感兴趣区域。
为了能够进一步有效提高对乳腺医学超声图像的特征识别的准确性及效率，在本申请的一实施例中，本申请还提供所述乳腺医学超声图像的自动化分类方法中步骤200的具体实施方式，参见图5，所述步骤200具体包括如下内容：
步骤201：对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的HSV颜色特征。
步骤202：在所述HSV颜色特征中提取所述目标乳腺医学超声图像对应的色调H通道值。
基于上述内容，本申请提供一种乳腺医学超声图像的自动化分类方法的模型预测场景，具体内容如下：
(1)获取目标乳腺弹性超声影像数据。
(2)对目标乳腺弹性超声影像数据进行数据清洗，删除一些系统设置参数等信息。
(3)对经数据清洗后的目标乳腺弹性超声影像数据中的目标区域进行标记，得到经数据清洗后的目标乳腺弹性超声影像数据的样本标签，其中，所述目标区域可以为医生对乳腺弹性超声影像数据中的感兴趣区域。
(4)将经数据清洗以及标记后的目标乳腺弹性超声影像数据进行HSV颜色空间转换，得到经数据清洗以及标记后的目标乳腺弹性超声影像数据对应的HSV颜色特征。
(5)在所述目标乳腺弹性超声影像数据对应的HSV颜色特征中，提取目标乳腺弹性超声影像数据对应的色调H通道值Hue value。
(6)将目标乳腺弹性超声影像数据对应的色调H通道值Hue value作为所述深度卷积神经网络的预测样本。
(7)将所述目标乳腺弹性超声影像数据对应的色调H通道值Hue value作为预测样本，输入所述深度卷积神经网络，并将该深度卷积神经网络的输出作为所述目标乳腺弹性超声影像数据的分类结果。
为了能够进一步有效提高乳腺医学超声图像的分类结果的准确性，在本申请的一实施例中，本申请还提供所述乳腺医学超声图像的自动化分类方法中在步骤100之前执行的模型建立及训练过程，参见图6，所述模型建立及训练过程具体包括如下内容：
步骤001：构建一深度卷积神经网络，以及，根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集。
步骤002：应用所述训练样本集对所述深度卷积神经网络进行模型训练。
其中，所述步骤001中的所述根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集，参见图7，具体包含有如下内容：
步骤001a：获取多个历史乳腺医学超声图像；
步骤002a：对各个历史乳腺医学超声图像均进行数据清洗；
步骤003a：对经数据清洗的各个历史乳腺医学超声图像中的目标区域分别进行标记，得到各个历史乳腺医学超声图像的样本标签；
步骤004a：将各个历史乳腺医学超声图像进行HSV颜色空间转换，得到各个历史乳腺医学超声图像分别对应的HSV颜色特征；
步骤005a：在各个历史乳腺医学超声图像的所述HSV颜色特征中，分别提取各个历史乳腺医学超声图像对应的色调H通道值；
步骤006a：将各个历史乳腺医学超声图像对应的色调H通道值组成所述深度卷积神经网络的训练样本集。
基于上述内容，本申请提供一种乳腺医学超声图像的自动化分类方法的模型建立及训练场景，具体内容如下：
(一)深度卷积神经网络建立
深度卷积神经网络，由多个卷积层和多个全连接层组成：输入图像在深度卷积神经网络中进行层层映射，得到各层对于图像不同的表示形式，实现图像的深度表示。其中，所述深度卷积神经网络中具体包含有5个卷积层：模块1至模块5，还具体包含有由3个全连接层组成的模块6，所述模块1至模块6依次排列。
所述深度卷积神经网络的具体结果如图8所示，各模块中的具体内容详见下表1。
表1
Figure PCTCN2018117460-appb-000001
用式(1)来表示卷积网络中前向传播各层网络的输出，即：
Figure PCTCN2018117460-appb-000002
式中，*代表各层输入与滤波器之间的卷积操作；
Figure PCTCN2018117460-appb-000003
是第l层卷积后第j个神经元的输出；
Figure PCTCN2018117460-appb-000004
是第l-1层第k个神经元的输出，即第l层的输入数据；
Figure PCTCN2018117460-appb-000005
是第l-1层第k个神经元连接到第l层第j个神经元的权重；c为第l层第j个神经元对应的偏置；f(·)是非线性函数，即激活函数，在本申请网络中间层使用ReLU(rectified linear unit)激活函数，其数学表达式f(x)＝max(0,x)。在实际应用中，还存在其他变换形式的ReLU函数，比如：Leaky-ReLU、P-ReLU等。
本申请所设计的深度卷积神经网络的最后几层通过全连接层构成，最后一层的激活函数为softmax函数，用于乳腺超声图像局部特征的分类，用式(2)表示如下：
C L＝softmax(W LC L-1+b L)           (2)
式中，
Figure PCTCN2018117460-appb-000006
且W L和b L分别为最后一层全连接层的权重和偏置。
可以理解的，所述获取训练样本的过程可以与建立深度卷积神经网络的过程并行执行，也可以在建立深度卷积神经网络的过程之前或之后执行。
(二)训练数据采集
获取大量的历史乳腺弹性超声影像数据。
(三)训练数据预处理
(1)对每个历史乳腺弹性超声影像数据均进行数据清洗，删除一些系统设置参数等信息。
(2)对经数据清洗后的各个历史乳腺弹性超声影像数据中的目标区域进行标记，得到经数据清洗后的各个历史乳腺弹性超声影像数据的样本标签，其中，所述目标区域可以为医生对乳腺弹性超声影像数据中的感兴趣区域。
(四)特征提取，训练样本集生成
(1)将经数据清洗以及标记后的各个历史乳腺弹性超声影像数据进行HSV颜色空间转换，得到经数据清洗以及标记后的各个历史乳腺弹性超声影像数据对应的HSV颜色特征。
(2)在各个历史乳腺弹性超声影像数据的所述HSV颜色特征中，分别提取各个历史乳腺弹性超声影像数据对应的色调H通道值Hue value：
Figure PCTCN2018117460-appb-000007
(3)将各个历史乳腺弹性超声影像数据对应的色调H通道值Hue value组成所述深度卷积神经网络的训练样本集。
(五)模型训练
将由各个所述历史乳腺弹性超声影像数据分别对应的色调H通道值Hue value组成的训练样本集，对所述深度卷积神经网络进行模型训练。
为了能够进一步有效提高乳腺医学超声图像的分类结果的准确性，在本申请的一实施例中，本申请还提供所述乳腺医学超声图像的自动化分类方法中在步骤100之前及步骤002之后执行的模型测试过程，参见图9，所述模型测试过程具体包括如下内容：
步骤A01：根据至少一个测试用乳腺医学超声图像得到测试样本；
步骤A02：应用所述测试样本对所述深度卷积神经网络进行模型测试，并将该深度卷积神经网络的输出作为测试结果；
步骤A03：基于所述测试结果及至少一个测试用乳腺医学超声图像的已知分类结果，判断当前深度卷积神经网络是否符合预设要求；
若是，则执行步骤A04，否则，则执行步骤A05。
步骤A04：将当前深度卷积神经网络作为用于对乳腺医学超声图像进行分类的目标深度卷积神经网络。
步骤A05：若当前深度卷积神经网络不符合所述预设要求，则对当前深度卷积神经网络进行优化和/或应用更新后的训练样本集重新对该深度卷积神经网络进行模型训练。
基于上述内容，本申请提供一种乳腺医学超声图像的自动化分类方法的模型测试场景，具体内容如下：
(1)获取未在历史乳腺弹性超声影像数据中出现的若干测试用乳腺弹性超声影像数据以及其已知的分类结果；
(2)对各个测试用乳腺弹性超声影像数据进行数据清洗，删除一些系统设置参数等信息。
(3)对各个经数据清洗后的测试用乳腺弹性超声影像数据中的目标区域进行标记，得到各个经数据清洗后的测试用乳腺弹性超声影像数据的样本标签，其中，所述目标区域可以为医生对乳腺弹性超声影像数据中的感兴趣区域。
(4)将各个经数据清洗以及标记后的测试用乳腺弹性超声影像数据进行HSV颜色空间转换，得到各个经数据清洗以及标记后的测试用乳腺弹性超声影像数据对应的HSV颜色特征。
(5)在各个所述测试用乳腺弹性超声影像数据对应的HSV颜色特征中，提取各个测试用乳腺弹性超声影像数据对应的色调H通道值Hue value。
(6)将各个测试用乳腺弹性超声影像数据对应的色调H通道值Hue value作为所述深度卷积神经网络的测试样本。
(7)将各个测试用乳腺弹性超声影像数据对应的色调H通道值Hue value作为测试样本，输入所述深度卷积神经网络，并将该深度卷积神经网络的输出的各个测试用乳腺弹性超声影像数据的分类结果与其对应的已知的分类结果进行比较，若出现不一致的情形，或者，出现不一致的次数超过预设值，则返回深度卷积神经网络建立的过程和/或训 练样本获取的过程进行优化，并在优化后进行重新测试，若新的测试结果符合预设要求，则确认该当前的深度卷积神经网络能够用于目标乳腺弹性超声影像数据的分类。
其中，返回深度卷积神经网络建立的过程对所述深度卷积神经网络进行优化的具体方式如下：
深度卷积神经网络后向传播过程通过BP反向传播算法实现权重和偏置的更新，通过优化分类交叉熵的损失函数实现深度卷积神经网络参数的最优化，如式(4)所示，使其能够提取到能量化乳腺超声图像中感兴趣区域的特征，达到对乳腺超声图像感兴趣区域特征进行快速、准确分类的目的。
Figure PCTCN2018117460-appb-000008
式中，I和K为分类类别数和总的训练样本数，Y′ i为健康区域或病变区域的标注。
从上述描述可知，本申请实施例提供的乳腺医学超声图像的自动化分类方法，可用于对乳腺感兴趣区域特征进行分类，比如健康区域或病变区域或可疑病变区域特征等，但不局限与乳腺位置。通过利用深度学习网络对乳腺超声图像进行训练，通过网络不断的自我学习和特征提取，达到对乳腺影像特征的自动分类目的。以及，本申请技术通过对乳腺医学影像数据进行分析，有助于减低临床医生工作量，提高医生工作效率；此外该方法能够以其无创、实时、安全和方便的优势在乳腺医学图像特征分类上进行应用。
为了能够有效提高对乳腺医学超声图像的特征识别的准确性及效率，并能够有效可靠提高对乳腺医学超声图像的分类过程的高效性及智能化程度，以及提高乳腺医学超声图像的分类结果的准确性，本申请实施例提供一种用于实现所述乳腺医学超声图像的自动化分类方法的全部内容的乳腺医学超声图像的自动化分类装置，参见图10，所述乳腺医学超声图像的自动化分类装置具体包括如下内容：
数据预处理模块10，用于对获取的目标乳腺医学超声图像进行预处理。
通道特征提取模块20，用于对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征。
模型预测模块30，用于将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
本申请提供的乳腺医学超声图像的自动化分类装置的实施例具体可以用于执行上述实施例中的乳腺医学超声图像的自动化分类方法的各个实施例的全部处理流程，其功能在此不再赘述，可以参照上述方法实施例的详细描述。
从上述描述可知，本申请实施例提供的乳腺医学超声图像的自动化分类装置，通过数据预处理模块10对获取的目标乳腺医学超声图像进行预处理，能够为后续图像识别提供准确且可靠的数据基础，通过通道特征提取模块20对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征，能够提取高通量、高层次影像特征用于乳腺超声图像特征分类，通过模型预测模块30将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果，能够有效提高对乳腺医学超声图像的特征识别的准确性及效率，并能够有效可靠提高对乳腺医学超声图像的分类过程的高效性及智能化程度，以及提高乳腺医学超声图像的分类结果的准确性，进而能够满足现代化乳腺医学的自动化需求并能够有效适应现在医学影像大数据智能诊断的发展趋势。
为了通过提高数据基础的准确性来进一步有效提高图像分类的准确性，在本申请的一实施例中，本申请还提供所述乳腺医学超声图像的自动化分类装置中数据预处理模块10的具体实施方式，所述数据预处理模块10具体包括如下内容：
目标数据获取单元11，用于获取目标乳腺医学超声图像。
目标数据清洗单元12，用于对所述目标乳腺医学超声图像进行数据清洗。
目标数据标记单元13，用于对经数据清洗后的目标乳腺医学超声图像中的目标区域进行标记，得到所述目标乳腺医学超声图像的样本标签。
为了能够进一步有效提高对乳腺医学超声图像的特征识别的准确性及效率，在本申请的一实施例中，本申请还提供所述乳腺医学超声图像的自动化分类装置中通道特征提取模块20的具体实施方式，所述通道特征提取模块20具体包括如下内容：
目标HSV转换单元21，用于对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的HSV颜色特征。
目标H通道提取单元22，用于在所述HSV颜色特征中提取所述目标乳腺医学超声图像对应的色调H通道值。
为了能够进一步有效提高乳腺医学超声图像的分类结果的准确性，在本申请的一实施例中，本申请还提供所述乳腺医学超声图像的自动化分类装置中的模型构建模块和模型训练模块，参见图11，所述模型构建模块和模型训练模块具体包括如下内容：
模型构建模块01，用于构建一深度卷积神经网络，以及，根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集。
模型训练模块02，用于应用所述训练样本集对所述深度卷积神经网络进行模型训练。
可以理解的是，所述深度卷积神经网络由多个卷积层和多个全连接层依次组成，乳腺医学超声图像经首个卷积层输入后，在所述深度卷积神经网络中进行层层映射，得到各层对于该乳腺医学超声图像不同的表示形式。
其中的模型构建模块01还具体包含有如下内容：
历史数据获取单元01a，用于获取多个历史乳腺医学超声图像；
历史数据清洗单元01b，用于对各个历史乳腺医学超声图像均进行数据清洗；
历史数据标记单元01c，用于对经数据清洗的各个历史乳腺医学超声图像中的目标区域分别进行标记，得到各个历史乳腺医学超声图像的样本标签；
历史HSV转换单元01d，用于将各个历史乳腺医学超声图像进行HSV颜色空间转换，得到各个历史乳腺医学超声图像分别对应的HSV颜色特征；
历史H通道提取单元01e，用于在各个历史乳腺医学超声图像的所述HSV颜色特征中，分别提取各个历史乳腺医学超声图像对应的色调H通道值；
训练样本集生成单元01f，用于将各个历史乳腺医学超声图像对应的色调H通道值组成所述深度卷积神经网络的训练样本集。
为了能够进一步有效提高乳腺医学超声图像的分类结果的准确性，在本申请的一实施例中，参见图12，本申请还提供所述乳腺医学超声图像的自动化分类装置中的模型测试模块A0，所述模型测试模块A0具体包括如下内容：
测试样本获取单元A1，用于根据至少一个测试用乳腺医学超声图像得到测试样本；
模型测试单元A2，用于应用所述测试样本对所述深度卷积神经网络进行模型测试，并将该深度卷积神经网络的输出作为测试结果；
测试结果判定单元A3，用于基于所述测试结果及至少一个测试用乳腺医学超声图像的已知分类结果，判断当前深度卷积神经网络是否符合预设要求，若是，则将当前深度卷积神经网络作为用于对乳腺医学超声图像进行分类的目标深度卷积神经网络。且所述测试结果判定单元还用于若当前深度卷积神经网络不符合所述预设要求，则对当前深度卷积神经网络进行优化和/或应用更新后的训练样本集重新对该深度卷积神经网络进行模型训练。
从上述描述可知，本申请实施例提供的乳腺医学超声图像的自动化分类装置，可用于对乳腺感兴趣区域特征进行分类，比如健康区域或病变区域或可疑病变区域特征等， 但不局限与乳腺位置。通过利用深度学习网络对乳腺超声图像进行训练，通过网络不断的自我学习和特征提取，达到对乳腺影像特征的自动分类目的。以及，本申请技术通过对乳腺医学影像数据进行分析，有助于减低临床医生工作量，提高医生工作效率；此外该装置能够以其无创、实时、安全和方便的优势在乳腺医学图像特征分类上进行应用。
本申请的实施例还提供能够实现上述实施例中的乳腺医学超声图像的自动化分类方法中全部步骤的一种电子设备的具体实施方式，参见图13，所述电子设备具体包括如下内容：
处理器(processor)601、存储器(memory)602、通信接口(Communications Interface)603和总线604；
其中，所述处理器601、存储器602、通信接口603通过所述总线604完成相互间的通信；所述通信接口603用于实现乳腺医学超声图像的自动化分类装置、服务器、客户终端以及其他参与机构之间的信息传输；
所述处理器601用于调用所述存储器602中的计算机程序，所述处理器执行所述计算机程序时实现上述实施例中的乳腺医学超声图像的自动化分类方法中的全部步骤，例如，所述处理器执行所述计算机程序时实现下述步骤：
步骤100：对获取的目标乳腺医学超声图像进行预处理。
步骤200：对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征。
步骤300：将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
从上述描述可知，本申请实施例提供的电子设备，可用于对乳腺感兴趣区域特征进行分类，比如健康区域或病变区域或可疑病变区域特征等，但不局限与乳腺位置。通过利用深度学习网络对乳腺超声图像进行训练，通过网络不断的自我学习和特征提取，达到对乳腺影像特征的自动分类目的。以及，本申请技术通过对乳腺医学影像数据进行分析，有助于减低临床医生工作量，提高医生工作效率；此外该装置能够以其无创、实时、安全和方便的优势在乳腺医学图像特征分类上进行应用。
本申请的实施例还提供能够实现上述实施例中的乳腺医学超声图像的自动化分类方法中全部步骤的一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机程 序，该计算机程序被处理器执行时实现上述实施例中的乳腺医学超声图像的自动化分类方法的全部步骤，例如，所述处理器执行所述计算机程序时实现下述步骤：
步骤100：对获取的目标乳腺医学超声图像进行预处理。
步骤200：对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征。
步骤300：将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
从上述描述可知，本申请实施例提供的计算机可读存储介质，可用于对乳腺感兴趣区域特征进行分类，比如健康区域或病变区域或可疑病变区域特征等，但不局限与乳腺位置。通过利用深度学习网络对乳腺超声图像进行训练，通过网络不断的自我学习和特征提取，达到对乳腺影像特征的自动分类目的。以及，本申请技术通过对乳腺医学影像数据进行分析，有助于减低临床医生工作量，提高医生工作效率；此外该装置能够以其无创、实时、安全和方便的优势在乳腺医学图像特征分类上进行应用。
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于硬件+程序类实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
上述对本说明书特定实施例进行了描述。其它实施例在所附权利要求书的范围内。在一些情况下，在权利要求书中记载的动作或步骤可以按照不同于实施例中的顺序来执行并且仍然可以实现期望的结果。另外，在附图中描绘的过程不一定要求示出的特定顺序或者连续顺序才能实现期望的结果。在某些实施方式中，多任务处理和并行处理也是可以的或者可能是有利的。
虽然本申请提供了如实施例或流程图所述的方法操作步骤，但基于常规或者无创造性的劳动可以包括更多或者更少的操作步骤。实施例中列举的步骤顺序仅仅为众多步骤执行顺序中的一种方式，不代表唯一的执行顺序。在实际中的装置或客户端产品执行时，可以按照实施例或者附图所示的方法顺序执行或者并行执行(例如并行处理器或者多线程处理的环境)。
上述实施例阐明的系统、装置、模块或单元，具体可以由计算机芯片或实体实现，或者由具有某种功能的产品来实现。一种典型的实现设备为计算机。具体的，计算机例 如可以为个人计算机、膝上型计算机、车载人机交互设备、蜂窝电话、相机电话、智能电话、个人数字助理、媒体播放器、导航设备、电子邮件设备、游戏控制台、平板计算机、可穿戴设备或者这些设备中的任何设备的组合。
虽然本说明书实施例提供了如实施例或流程图所述的方法操作步骤，但基于常规或者无创造性的手段可以包括更多或者更少的操作步骤。实施例中列举的步骤顺序仅仅为众多步骤执行顺序中的一种方式，不代表唯一的执行顺序。在实际中的装置或终端产品执行时，可以按照实施例或者附图所示的方法顺序执行或者并行执行(例如并行处理器或者多线程处理的环境，甚至为分布式数据处理环境)。术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、产品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、产品或者设备所固有的要素。在没有更多限制的情况下，并不排除在包括所述要素的过程、方法、产品或者设备中还存在另外的相同或等同要素。
为了描述的方便，描述以上装置时以功能分为各种模块分别描述。当然，在实施本说明书实施例时可以把各模块的功能在同一个或多个软件和/或硬件中实现，也可以将实现同一功能的模块由多个子模块或子单元的组合实现等。以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
本领域技术人员也知道，除了以纯计算机可读程序代码方式实现控制器以外，完全可以通过将方法步骤进行逻辑编程来使得控制器以逻辑门、开关、专用集成电路、可编程逻辑控制器和嵌入微控制器等的形式来实现相同功能。因此这种控制器可以被认为是一种硬件部件，而对其内部包括的用于实现各种功能的装置也可以视为硬件部件内的结构。或者甚至，可以将用于实现各种功能的装置视为既可以是实现方法的软件模块又可以是硬件部件内的结构。
本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备 的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
本领域技术人员应明白，本说明书的实施例可提供为方法、系统或计算机程序产品。因此，本说明书实施例可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本说明书实施例可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本说明书实施例可以在由计算机执行的计算机可执行指令的一般上下文中描述，例如程序模块。一般地，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中实践本说明书实施例，在这些分布式计算环境中，由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中，程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于系统实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。在本说明书的描述中，参考术语“一个实施例”、“一些实施例”、“示例”、“具体示例”、或“一些示例”等的描述意指结合该实施例或示例描述的具体特征、结构、材料或者特点包含于本说明书实施例的至少一个实施例或示例中。在本说明书中，对上述术语的示意性表述不必须针对的是相同的实施例或示例。而且，描述的具体特征、结构、材料或者特点可以在任一个或多个实施例或示例中以合适的方式结合。此外，在不相互矛盾的情况下，本领域的技术人员可以将本说明书中描述的不同实施例或示例以及不同实施例或示例的特征进行结合和组合。
以上所述仅为本说明书实施例的实施例而已，并不用于限制本说明书实施例。对于本领域技术人员来说，本说明书实施例可以有各种更改和变化。凡在本说明书实施例的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本说明书实施例的权利要求范围之内。

Claims (18)

  1. 一种乳腺医学超声图像的自动化分类方法，其特征在于，包括：
    对获取的目标乳腺医学超声图像进行预处理；
    对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征；
    将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
  2. 根据权利要求1所述的乳腺医学超声图像的自动化分类方法，其特征在于，所述对获取的目标乳腺医学超声图像进行预处理，包括：
    获取目标乳腺医学超声图像；
    对所述目标乳腺医学超声图像进行数据清洗；
    对经数据清洗后的目标乳腺医学超声图像中的目标区域进行标记，得到所述目标乳腺医学超声图像的样本标签。
  3. 根据权利要求1所述的乳腺医学超声图像的自动化分类方法，其特征在于，所述对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征，包括：
    对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的HSV颜色特征；
    在所述HSV颜色特征中提取所述目标乳腺医学超声图像对应的色调H通道值。
  4. 根据权利要求1所述的乳腺医学超声图像的自动化分类方法，其特征在于，在所述对获取的目标乳腺医学超声图像进行预处理之前，还包括：
    构建一深度卷积神经网络，以及，根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集；
    应用所述训练样本集对所述深度卷积神经网络进行模型训练。
  5. 根据权利要求4所述的乳腺医学超声图像的自动化分类方法，其特征在于，所述深度卷积神经网络由多个卷积层和多个全连接层依次组成，乳腺医学超声图像经首个卷积层输入后，在所述深度卷积神经网络中进行层层映射，得到各层对于该乳腺医学超声图像不同的表示形式。
  6. 根据权利要求4所述的乳腺医学超声图像的自动化分类方法，其特征在于，所述根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集，包括：
    获取多个历史乳腺医学超声图像；
    对各个历史乳腺医学超声图像均进行数据清洗；
    对经数据清洗的各个历史乳腺医学超声图像中的目标区域分别进行标记，得到各个历史乳腺医学超声图像的样本标签；
    将各个历史乳腺医学超声图像进行HSV颜色空间转换，得到各个历史乳腺医学超声图像分别对应的HSV颜色特征；
    在各个历史乳腺医学超声图像的所述HSV颜色特征中，分别提取各个历史乳腺医学超声图像对应的色调H通道值；
    将各个历史乳腺医学超声图像对应的色调H通道值组成所述深度卷积神经网络的训练样本集。
  7. 根据权利要求4所述的乳腺医学超声图像的自动化分类方法，其特征在于，还包括：
    根据至少一个测试用乳腺医学超声图像得到测试样本；
    应用所述测试样本对所述深度卷积神经网络进行模型测试，并将该深度卷积神经网络的输出作为测试结果；
    基于所述测试结果及至少一个测试用乳腺医学超声图像的已知分类结果，判断当前深度卷积神经网络是否符合预设要求；
    若是，则将当前深度卷积神经网络作为用于对乳腺医学超声图像进行分类的目标深度卷积神经网络。
  8. 根据权利要求7所述的乳腺医学超声图像的自动化分类方法，其特征在于，还包括：
    若当前深度卷积神经网络不符合所述预设要求，则对当前深度卷积神经网络进行优化和/或应用更新后的训练样本集重新对该深度卷积神经网络进行模型训练。
  9. 一种乳腺医学超声图像的自动化分类装置，其特征在于，包括：
    数据预处理模块，用于对获取的目标乳腺医学超声图像进行预处理；
    通道特征提取模块，用于对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的色调H通道特征；
    模型预测模块，用于将预处理后的所述目标乳腺医学超声图像对应的色调H通道特征作为预测样本，输入预设的目标深度卷积神经网络，并将该目标深度卷积神经网络的输出作为所述目标乳腺医学超声图像的分类结果。
  10. 根据权利要求9所述的乳腺医学超声图像的自动化分类装置，其特征在于，所述数据预处理模块，包括：
    目标数据获取单元，用于获取目标乳腺医学超声图像；
    目标数据清洗单元，用于对所述目标乳腺医学超声图像进行数据清洗；
    目标数据标记单元，用于对经数据清洗后的目标乳腺医学超声图像中的目标区域进行标记，得到所述目标乳腺医学超声图像的样本标签。
  11. 根据权利要求9所述的乳腺医学超声图像的自动化分类装置，其特征在于，所述通道特征提取模块包括：
    目标HSV转换单元，用于对预处理后的所述目标乳腺医学超声图像进行HSV颜色空间转换，得到该目标乳腺医学超声图像对应的HSV颜色特征；
    目标H通道提取单元，用于在所述HSV颜色特征中提取所述目标乳腺医学超声图像对应的色调H通道值。
  12. 根据权利要求9所述的乳腺医学超声图像的自动化分类装置，其特征在于，还包括：
    模型构建模块，用于构建一深度卷积神经网络，以及，根据多个历史乳腺医学超声图像获取该深度卷积神经网络的训练样本集；
    模型训练模块，用于应用所述训练样本集对所述深度卷积神经网络进行模型训练。
  13. 根据权利要求12所述的乳腺医学超声图像的自动化分类装置，其特征在于，所述深度卷积神经网络由多个卷积层和多个全连接层依次组成，乳腺医学超声图像经首个卷积层输入后，在所述深度卷积神经网络中进行层层映射，得到各层对于该乳腺医学超声图像不同的表示形式。
  14. 根据权利要求12所述的乳腺医学超声图像的自动化分类装置，其特征在于，所述模型构建模块包括：
    历史数据获取单元，用于获取多个历史乳腺医学超声图像；
    历史数据清洗单元，用于对各个历史乳腺医学超声图像均进行数据清洗；
    历史数据标记单元，用于对经数据清洗的各个历史乳腺医学超声图像中的目标区域分别进行标记，得到各个历史乳腺医学超声图像的样本标签；
    历史HSV转换单元，用于将各个历史乳腺医学超声图像进行HSV颜色空间转换，得到各个历史乳腺医学超声图像分别对应的HSV颜色特征；
    历史H通道提取单元，用于在各个历史乳腺医学超声图像的所述HSV颜色特征中，分别提取各个历史乳腺医学超声图像对应的色调H通道值；
    训练样本集生成单元，用于将各个历史乳腺医学超声图像对应的色调H通道值组成所述深度卷积神经网络的训练样本集。
  15. 根据权利要求12所述的乳腺医学超声图像的自动化分类装置，其特征在于，还包括：
    测试样本获取单元，用于根据至少一个测试用乳腺医学超声图像得到测试样本；
    模型测试单元，用于应用所述测试样本对所述深度卷积神经网络进行模型测试，并将该深度卷积神经网络的输出作为测试结果；
    测试结果判定单元，用于基于所述测试结果及至少一个测试用乳腺医学超声图像的已知分类结果，判断当前深度卷积神经网络是否符合预设要求，若是，则将当前深度卷积神经网络作为用于对乳腺医学超声图像进行分类的目标深度卷积神经网络。
  16. 根据权利要求15所述的乳腺医学超声图像的自动化分类装置，其特征在于，所述测试结果判定单元还用于若当前深度卷积神经网络不符合所述预设要求，则对当前深度卷积神经网络进行优化和/或应用更新后的训练样本集重新对该深度卷积神经网络进行模型训练。
  17. 一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现如权利要求1至8任一项所述的乳腺医学超声图像的自动化分类方法的步骤。
  18. 一种计算机可读存储介质，其上存储有计算机程序，其特征在于，该计算机程序被处理器执行时实现如权利要求1至8任一项所述的乳腺医学超声图像的自动化分类方法的步骤。
PCT/CN2018/117460 2018-11-26 2018-11-26 乳腺医学超声图像的自动化分类方法及装置 Ceased WO2020107156A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/117460 WO2020107156A1 (zh) 2018-11-26 2018-11-26 乳腺医学超声图像的自动化分类方法及装置

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/117460 WO2020107156A1 (zh) 2018-11-26 2018-11-26 乳腺医学超声图像的自动化分类方法及装置

Publications (1)

Publication Number Publication Date
WO2020107156A1 true WO2020107156A1 (zh) 2020-06-04

Family

ID=70852458

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/117460 Ceased WO2020107156A1 (zh) 2018-11-26 2018-11-26 乳腺医学超声图像的自动化分类方法及装置

Country Status (1)

Country Link
WO (1) WO2020107156A1 (zh)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111862015A (zh) * 2020-07-08 2020-10-30 中国人民解放军战略支援部队信息工程大学 一种图像质量等级确定方法、装置及电子设备
CN112699948A (zh) * 2020-12-31 2021-04-23 无锡祥生医疗科技股份有限公司 超声乳腺病灶的分类方法、装置及存储介质
CN113139076A (zh) * 2021-05-20 2021-07-20 广东工业大学 一种深度特征学习多标签的神经网络影像自动标记方法
CN113421633A (zh) * 2021-06-25 2021-09-21 上海联影智能医疗科技有限公司 特征分类方法、计算机设备和存储介质
CN113744261A (zh) * 2021-09-17 2021-12-03 天津市肿瘤医院（天津医科大学肿瘤医院） 基于卷积神经网络的甲状腺超声图像的特征提取方法
CN114004798A (zh) * 2021-10-27 2022-02-01 深圳大学 一种多模态乳腺结节的良恶性预测方法及相关装置
CN115019046A (zh) * 2022-07-01 2022-09-06 深圳大学 甲状腺超声图像自动识别方法、装置、终端及存储介质
CN115564763A (zh) * 2022-11-01 2023-01-03 上海杏脉信息科技有限公司 甲状腺超声图像处理方法、装置、介质及电子设备
CN116309585A (zh) * 2023-05-22 2023-06-23 山东大学 基于多任务学习的乳腺超声图像目标区域识别方法及系统
CN118397380A (zh) * 2024-06-28 2024-07-26 北京大学第三医院（北京大学第三临床医学院） 甲状腺结节性病变分类模型训练方法、分类方法及装置

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080212887A1 (en) * 2005-05-12 2008-09-04 Bracco Imaging S.P.A. Method For Coding Pixels or Voxels of a Digital Image and a Method For Processing Digital Images
CN106339591A (zh) * 2016-08-25 2017-01-18 汤平 一种基于深度卷积神经网络的预防乳腺癌自助健康云服务系统
CN107028593A (zh) * 2017-04-14 2017-08-11 成都知识视觉科技有限公司 一种乳腺导管原位癌的辅助检测方法
CN107705336A (zh) * 2017-04-15 2018-02-16 北京航空航天大学 一种病理图像染色成分调节方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20080212887A1 (en) * 2005-05-12 2008-09-04 Bracco Imaging S.P.A. Method For Coding Pixels or Voxels of a Digital Image and a Method For Processing Digital Images
CN106339591A (zh) * 2016-08-25 2017-01-18 汤平 一种基于深度卷积神经网络的预防乳腺癌自助健康云服务系统
CN107028593A (zh) * 2017-04-14 2017-08-11 成都知识视觉科技有限公司 一种乳腺导管原位癌的辅助检测方法
CN107705336A (zh) * 2017-04-15 2018-02-16 北京航空航天大学 一种病理图像染色成分调节方法

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111862015B (zh) * 2020-07-08 2024-03-19 中国人民解放军战略支援部队信息工程大学 一种图像质量等级确定方法、装置及电子设备
CN111862015A (zh) * 2020-07-08 2020-10-30 中国人民解放军战略支援部队信息工程大学 一种图像质量等级确定方法、装置及电子设备
CN112699948A (zh) * 2020-12-31 2021-04-23 无锡祥生医疗科技股份有限公司 超声乳腺病灶的分类方法、装置及存储介质
CN113139076A (zh) * 2021-05-20 2021-07-20 广东工业大学 一种深度特征学习多标签的神经网络影像自动标记方法
CN113139076B (zh) * 2021-05-20 2024-03-29 广东工业大学 一种深度特征学习多标签的神经网络影像自动标记方法
CN113421633A (zh) * 2021-06-25 2021-09-21 上海联影智能医疗科技有限公司 特征分类方法、计算机设备和存储介质
CN113744261A (zh) * 2021-09-17 2021-12-03 天津市肿瘤医院（天津医科大学肿瘤医院） 基于卷积神经网络的甲状腺超声图像的特征提取方法
CN114004798A (zh) * 2021-10-27 2022-02-01 深圳大学 一种多模态乳腺结节的良恶性预测方法及相关装置
CN115019046A (zh) * 2022-07-01 2022-09-06 深圳大学 甲状腺超声图像自动识别方法、装置、终端及存储介质
CN115564763A (zh) * 2022-11-01 2023-01-03 上海杏脉信息科技有限公司 甲状腺超声图像处理方法、装置、介质及电子设备
CN116309585B (zh) * 2023-05-22 2023-08-22 山东大学 基于多任务学习的乳腺超声图像目标区域识别方法及系统
CN116309585A (zh) * 2023-05-22 2023-06-23 山东大学 基于多任务学习的乳腺超声图像目标区域识别方法及系统
CN118397380A (zh) * 2024-06-28 2024-07-26 北京大学第三医院（北京大学第三临床医学院） 甲状腺结节性病变分类模型训练方法、分类方法及装置

Similar Documents

Publication Publication Date Title
WO2020107156A1 (zh) 乳腺医学超声图像的自动化分类方法及装置
CN110689025B (zh) 图像识别方法、装置、系统及内窥镜图像识别方法、装置
Remya et al. A novel transfer learning framework for multimodal skin lesion analysis
CN108197664A (zh) 模型获取方法、装置、电子设备及计算机可读存储介质
CN116664930A (zh) 基于自监督对比学习的个性化联邦学习图像分类方法及系统
WO2020118618A1 (zh) 一种乳腺肿块图像识别方法及装置
WO2021098534A1 (zh) 相似度确定、网络训练、查找方法及装置、电子装置和存储介质
CN109614993A (zh) 乳腺医学超声图像的自动化分类方法及装置
JP7225731B2 (ja) 多変数データシーケンスの画像化
CN119670807B (zh) 专家知识驱动的大模型定制化数据处理方法及相关设备
CN116188501A (zh) 基于多尺度交叉注意力的医学图像分割方法
CN118865477A (zh) 基于知识蒸馏和自监督学习的近视性黄斑病变分类方法
Li et al. Cytopathology image analysis method based on high-resolution medical representation learning in medical decision-making system
CN118761985A (zh) 一种基于图像分类模型及视觉处理的皮肤状况评测方法
CN116259311B (zh) 一种语音处理方法及其相关设备
CN109636780A (zh) 乳腺密度自动分级方法及装置
WO2024045819A1 (zh) 病变区域确定方法、模型训练方法及装置
CN117061733A (zh) 一种视频评估方法及其相关设备
CN116168245A (zh) 基于CvT-13和多模态图像融和的心电分类方法
CN120431068A (zh) 基于计算机视觉的血细胞分析系统及方法
CN117611910B (zh) 基于食管内镜图像的自监督识别方法、系统、设备及介质
CN119107322A (zh) 一种轻量级图像分割方法、装置、计算机设备及存储介质
Nooralli et al. Predictive Analytics for Tongue Diseases: A Comparative Study of Deep Learning Models
Ahmed et al. Melanoma Detection Using Augmented ResNet34 for High-Precision Dermoscopic Image Classification
CN113807263B (zh) 一种图像中字段的处理方法及其相关设备

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: 18941320

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

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 1205A DATED 08/11/2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18941320

Country of ref document: EP

Kind code of ref document: A1