CN112766332A - Medical image detection model training method, medical image detection method and device - Google Patents

Medical image detection model training method, medical image detection method and device Download PDF

Info

Publication number
CN112766332A
CN112766332A CN202110023647.6A CN202110023647A CN112766332A CN 112766332 A CN112766332 A CN 112766332A CN 202110023647 A CN202110023647 A CN 202110023647A CN 112766332 A CN112766332 A CN 112766332A
Authority
CN
China
Prior art keywords
medical image
neural network
image detection
medical
prior
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.)
Pending
Application number
CN202110023647.6A
Other languages
Chinese (zh)
Inventor
潘丹
曾安
魏敢
容华斌
谢锐伟
蔡重芪
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.)
Guangdong Zhongke Tianji Medical Equipment Co Ltd
Original Assignee
Guangdong Zhongke Tianji Medical Equipment Co Ltd
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 Guangdong Zhongke Tianji Medical Equipment Co Ltd filed Critical Guangdong Zhongke Tianji Medical Equipment Co Ltd
Priority to CN202110023647.6A priority Critical patent/CN112766332A/en
Publication of CN112766332A publication Critical patent/CN112766332A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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
    • 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
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing

Landscapes

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

Abstract

The invention relates to a medical image detection model training method, a medical image detection method and a medical image detection device, wherein a priori medical image data set is obtained, and a neural network is established according to the priori medical image data set so as to train a neural network segmentation model; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the multi-scale convolution layer and residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of the medical care work.

Description

Medical image detection model training method, medical image detection method and device
Technical Field
The invention relates to the technical field of image processing, in particular to a medical image detection model training method, a medical image detection method and a medical image detection device.
Background
Medical imaging refers to the technique and process of obtaining images of internal tissues of a human body or a part of the human body in a non-invasive manner for medical treatment or medical research. Generally, medical images contain the following two relatively independent directions of study: medical imaging systems and medical image processing. After the imaging system in the hospital completes the imaging of the image, the image needs to be further processed, so as to facilitate the subsequent related processing such as medical diagnosis or data processing.
At present, the conventional medical image processing mainly aims at the processing of image characteristics, such as image enhancement on medical images for medical staff to observe or data compression on medical images for cross-domain transmission. Medical image processing greatly facilitates the development complexity of medical care work on an image level. However, in the imaging process of medical images, a great number of objective interference factors are naturally caused by individual differences of human bodies. Taking chest X-ray imaging of the lung as an example, due to individual differences of imaging objects, the acquired lung images are different in size, and segmentation detection may be inaccurate due to individual differences when the lung images are segmented by using the existing template for detection. Therefore, after the imaging of the medical image is completed, the following detection of the medical image still has the above defects.
Disclosure of Invention
Therefore, it is necessary to provide a medical image detection model training method, a medical image detection method and a medical image detection device for the defects still existing in the conventional medical image detection.
A medical image detection model training method comprises the following steps:
acquiring a prior medical image to obtain a prior medical image data set;
establishing a neural network according to the prior medical image data set subjected to medical information classification and labeling so as to train a neural network segmentation model;
and adding a multi-scale convolution layer and residual connection into a convolution network layer of the neural network segmentation model to obtain the medical image detection model.
According to the medical image detection model training method, a priori medical image data set is obtained, a neural network is established according to the priori medical image data set, and a neural network segmentation model is trained; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the multi-scale convolution layer and residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of the medical care work.
In one embodiment, before the process of acquiring the prior medical image, the method further comprises the following steps:
and carrying out image preprocessing on the prior medical image.
In one embodiment, a process for acquiring a prior medical image to obtain a prior medical image data set includes the steps of:
segmenting a prior region image of a prior medical image;
and acquiring a prior medical image data set according to the prior region image.
In one embodiment, the process of establishing a neural network according to the prior medical image data set after information classification and labeling includes the steps of:
scaling the prior medical image in the prior medical image data set;
inputting the scaled prior medical image into a neural network, performing convolution processing on a convolution layer in the neural network, and calculating a characteristic subgraph;
and extracting features according to the feature subgraph by an interested region pooling layer in the neural network so as to carry out full-connection processing and classification processing of the logistic regression network.
In one embodiment, the loss function of the logistic regression network is as follows:
Figure BDA0002889451500000031
wherein i represents anchors index, piRepresenting the forkround softmax mobility,
Figure BDA0002889451500000032
representing the corresponding GT prediction probability, i.e. IoU between the ith anchor and GT>0.7, the anchor is considered to be forego,
Figure BDA0002889451500000033
otherwise IoU<When 0.3, the anchor is considered to be background,
Figure BDA0002889451500000034
as for those 0.3<IoU<0.7 anchor does not participate in training; t represents a predictbounding box,
Figure BDA0002889451500000035
represents the GTbox corresponding to the forego anchor; the whole Loss is divided into the following two parts:
the cls loss, namely the softmax loss calculated by rpn _ cls _ loss layer is used for network training for classifying anchors as forward and background;
the reg loss, namely soomth L1 loss calculated by rpn _ loss _ bbox layer, is used for bounding box regression network training; wherein, in reg loss, multiply
Figure BDA0002889451500000036
Shows regression with only forego anchors in mind.
In one embodiment, the multi-scale convolution operation of the multi-scale convolution layer is as follows:
Figure BDA0002889451500000037
wherein the content of the first and second substances,
Figure BDA0002889451500000038
j-th output feature map of 1 x 1 convolution kernel connected with l-1 layers
Figure BDA0002889451500000039
And ith output feature map f of multi-scale convolution layeri 1*1，
Figure BDA00028894515000000311
For offset, n is the number of l-1 layer output profiles, ReLU () is the activation function, fi 3*3、fi 5 *5The same is true.
In one embodiment, the fill parameter of the 1 x 1 convolution kernel is 0, the fill parameter of the 3 x 3 convolution kernel is 1, and the fill parameter of the 5 x 5 convolution kernel is 2;
the output characteristic diagram of the multi-scale convolutional layer is as follows:
fi Multi_Scale＝Concat(fi 1*1，fi 3*3，fi 5*5)。
in one embodiment, the loss function corresponding to the residual join is a cross-entropy loss function, as follows:
Figure BDA0002889451500000041
wherein K is the number of species; y is a label, if the sample class is i, then yi1, otherwise equal to 0; p is a radical ofiIs the probability of class i being output by the classification model.
A medical image detection model training device comprises:
the data set determining module is used for acquiring a prior medical image to obtain a prior medical image data set;
the neural network establishing module is used for establishing a neural network according to the prior medical image data set subjected to medical information classification and labeling so as to train a neural network segmentation model;
and the model determining module is used for adding a multi-scale convolution layer and residual connection into the convolution network layer of the neural network segmentation model to obtain the medical image detection model.
The medical image detection model training device obtains a prior medical image data set, and establishes a neural network according to the prior medical image data set so as to train a neural network segmentation model; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
A computer storage medium having computer instructions stored thereon, the computer instructions when executed by a processor implement the medical image detection model training method of any of the above embodiments.
The computer storage medium obtains a prior medical image data set, and establishes a neural network according to the prior medical image data set so as to train a neural network segmentation model; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the medical image detection model training method according to any of the above embodiments when executing the computer program.
The computer equipment obtains a prior medical image data set, and establishes a neural network according to the prior medical image data set so as to train a neural network segmentation model; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
A medical image detection method comprises the following steps:
acquiring a medical image to be processed;
and inputting the medical image to be processed into the medical image detection model to obtain a medical image detection result.
According to the medical image segmentation method, after the medical image to be processed is obtained, the medical image to be processed is input into the medical image detection model, so that a medical image detection result is obtained, and interference information is removed for diagnosis of medical staff. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
A medical image detection apparatus, comprising:
the image acquisition module is used for acquiring a medical image to be processed;
and the image detection module is used for inputting the medical image to be processed into the medical image detection model so as to obtain a medical image detection result.
According to the medical image detection device, when the medical image to be processed is obtained, the medical image to be processed is input into the medical image detection model, so that a medical image detection result is obtained, and interference information is removed for diagnosis of medical staff. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
A computer storage medium having computer instructions stored thereon, the computer instructions when executed by a processor implement the medical image detection method of any of the above embodiments.
The computer storage medium inputs the medical image to be processed into the medical image detection model after the medical image to be processed is obtained, so that a medical image detection result is obtained, and interference information is removed for diagnosis of medical staff. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the medical image detection method according to any of the embodiments.
According to the computer equipment, when the medical image to be processed is obtained, the medical image to be processed is input into the medical image detection model, so that a medical image detection result is obtained, and interference information is removed for diagnosis of medical staff. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
Drawings
FIG. 1 is a flowchart of a medical image detection model training method according to an embodiment;
FIG. 2 is a flowchart illustrating a method for training a medical image inspection model according to another embodiment;
FIG. 3 is a flowchart illustrating a method for training a medical image inspection model according to yet another embodiment;
FIG. 4 is a schematic view of mask segmentation;
FIG. 5 is a schematic diagram of a model processing architecture;
FIG. 6 is a diagram illustrating multi-scale convolution and residual concatenation;
FIG. 7 is a block diagram of a medical image inspection model training apparatus according to an embodiment;
FIG. 8 is a flowchart of a medical image inspection method according to an embodiment;
fig. 9 is a block diagram of a medical image detection apparatus according to an embodiment.
Detailed Description
For better understanding of the objects, technical solutions and effects of the present invention, the present invention will be further explained with reference to the accompanying drawings and examples. Meanwhile, the following described examples are only for explaining the present invention, and are not intended to limit the present invention.
The embodiment of the invention provides a medical image detection model training method.
Fig. 1 is a flowchart illustrating a medical image detection model training method according to an embodiment, and as shown in fig. 1, the medical image detection model training method according to an embodiment includes steps S100 to S102:
s100, acquiring a prior medical image to obtain a prior medical image data set;
in the medical care developing process, a medical image exists, and the part of stored historical medical images, namely prior hospital images, are called. The medical image includes an X-ray image, a CT image, an ultrasound image, or the like. In order to better explain the technical solution of the embodiment of the present invention, a chest X-ray film of the lung is taken as the medical image in the embodiment for explanation.
Meanwhile, the determined prior medical image data set comprises a plurality of prior medical images, and the prior medical image data set comprises a training set and a testing set constructed by the prior medical images. In one embodiment, the prior medical image data set includes a plurality of data set forms including a chestX-ray data set.
In one embodiment, fig. 2 is a flowchart of a training method for medical image detection models according to another embodiment, and as shown in fig. 2, before the process of acquiring the prior medical image in step S100, the method further includes step S200:
and S200, carrying out image preprocessing on the prior medical image.
The image preprocessing is carried out on the prior medical image, so that the diversity of the prior medical image is improved, and the generalization performance of a subsequent model is improved. The image preprocessing comprises modes of rotation, translation, deformation, noise addition and the like.
In one embodiment, fig. 3 is a flowchart of a medical image detection model training method according to yet another embodiment, and as shown in fig. 3, the process of performing image preprocessing on the prior medical image in step S200 includes steps S300 and S301:
s300, carrying out image rotation and image turnover processing on the prior medical image so as to expand a prior medical image data set;
the priori medical image is rotated by taking 0 degrees, 90 degrees, 180 degrees and 270 degrees as rotating axes, and then horizontal overturning, vertical overturning and mirror image overturning are carried out to obtain each image after image rotation and image overturning processing, so that the extension of the priori medical image data set is realized.
S301, noise disturbance is added to the prior medical image after image rotation and image inversion processing.
The noise disturbance is added to the prior medical image to inhibit the high-frequency characteristics of the image, weaken the influence of the image on a medical image segmentation model and improve the learning capacity of the model. In one embodiment, the noise disturbance addition process is implemented by adding gaussian noise to fill the entire prior medical image with random white or black pixels.
In one embodiment, as shown in fig. 2, the process of acquiring a prior medical image in step S100 to obtain a prior medical image data set includes steps S400 and S401:
s400, segmenting a prior region image of a prior medical image;
s401, acquiring a priori medical image data set according to the priori region image.
The priori region image of the prior medical image can be segmented through the mask, and the prior region is generally a region concerned by medical staff in pathological diagnosis. Fig. 4 is a schematic diagram of mask segmentation, and as shown in fig. 4, taking an X-ray chest image as an example, a conventional X-ray chest image includes portions of the thorax, which are likely to cause erroneous judgment of lung information such as lung nodules when entering the model. And extracting images of the left and right lung lobe areas through a mask to serve as prior area images so as to neglect the thoracic cavity and other noise parts and establish a prior medical image data set.
S101, establishing a neural network according to the prior medical image data set subjected to medical information classification and labeling so as to train a neural network segmentation model;
the medical information is the representation of the prior medical image, including pathology, diagnosis and the like. In history prior, medical care work determines corresponding medical information according to prior medical images, and classification labels of the medical care work are used for representing classification of the medical information. Taking the prior medical image as a chest X-ray film of the lung as an example, the medical information comprises lung disease types, each segmented chest X-ray film corresponds to one or more lung disease types, and a label is configured for each type of lung disease, namely, the prior medical image is labeled with a medical information classification label. In one embodiment, a prior medical image may be labeled with one or more medical information classification labels.
By adding the multi-scale convolution layer and the residual connection, the characteristic diagram of the medical image detection model can obtain information of different scales, and the problem of detection performance reduction caused by too large area difference of a focus area is solved.
In one embodiment, as shown in fig. 2, the process of establishing a neural network according to the prior medical image data set after medical information classification and labeling in step S101 includes steps S500 to S502:
s500, scaling the prior medical image in the prior medical image data set;
fig. 5 is a schematic diagram of a model processing structure, and as shown in fig. 5, the masked prior medical image with size P × Q is scaled to a fixed size M × N, and the M × N image is sent to a neural network.
S501, inputting the zoomed prior medical image into a neural network, performing convolution processing on a convolution layer in the neural network, and calculating a characteristic subgraph;
as shown in fig. 5, convolutional layers Conv layers of the neural network are mainly composed of a residual module respnetblock, where respnetblock is composed of a multi-scale convolution, a reduced linear rectification function (lu), a Pooling layer, and a residual connection; the RPN (Region-generated Network) Network is firstly convoluted by 3 to 3, and then forms the foreground anchors and bounding box regression offsets respectively, and calculates the characteristic subgraphs based on the former and the later.
And S502, extracting features according to the feature subgraph through the interested region pooling layer in the neural network so as to perform full-connection processing and classification processing of a logistic regression network.
As shown in fig. 5, the Roi (regions of interest) stacking layer extracts the generic features and sends them to the subsequent fully-connected and logistic regression network (softmax) for classification (classification) by using the feature subgraphs and feature maps.
In one embodiment, the loss function of the logistic regression network is as follows:
Figure BDA0002889451500000111
wherein i represents anchors index, piRepresenting the forkround softmax mobility,
Figure BDA0002889451500000112
representing the corresponding GT prediction probability, i.e. IoU between the ith anchor and GT>0.7, the anchor is considered to be forego,
Figure BDA0002889451500000113
otherwise IoU<When 0.3, the anchor is considered to be background,
Figure BDA0002889451500000114
as for those 0.3<IoU<0.7 anchor does not participate in training; t represents a predictbounding box,
Figure BDA0002889451500000115
represents the GTbox corresponding to the forego anchor; the whole Loss is divided into the following two parts:
the cls loss, namely the softmax loss calculated by rpn _ cls _ loss layer is used for network training for classifying anchors as forward and background;
the reg loss, namely soomth L1 loss calculated by rpn _ loss _ bbox layer, is used for bounding box regression network training; wherein, in reg loss, multiply
Figure BDA0002889451500000116
Shows regression with only forego anchors in mind.
And S102, adding a multi-scale convolution layer and residual connection into a convolution network layer of the neural network segmentation model to obtain the medical image detection model.
In one embodiment, the neural network is a fast-RCNN network.
Fig. 6 is a schematic diagram of multi-scale convolution and residual connection, and as shown in fig. 6, a multi-scale convolution layer and residual connection are added to a convolution network layer of a neural network segmentation model, so that a feature map can obtain information of different scales, and the problem of detection performance degradation caused by too large area difference of a focus region is solved
In one embodiment, the multi-scale convolution operation of the multi-scale convolution layer is as follows:
Figure BDA0002889451500000121
wherein the content of the first and second substances,
Figure BDA0002889451500000123
j-th output feature map of 1 x 1 convolution kernel connected with l-1 layers
Figure BDA0002889451500000124
And ith output feature map f of multi-scale convolution layeri 1*1，
Figure BDA0002889451500000125
For offset, n is the number of l-1 layer output profiles, ReLU () is the activation function, fi 3*3、fi 5*5The same is true.
As a preferred embodiment, in order to make the characteristic diagram fi 1*1、fi 3*3And fi 5*5The sizes are the same, and filling parameters pad and pad are respectively set for convolution kernels with 3 different sizes1*1＝0、pad:pad3*3＝1、pad:pad5*5Finally, the feature map f is mapped by Concati 1*1、fi 3*3And fi 5*5Connecting to obtain a characteristic diagram f finally output by the multi-scale convolution layeri Multi _ScaleThe operation is represented by the following formula:
fi Multi_Scale＝Concat(fi 1*1，fi 3*3，fi 5*5)
when the neural network predicts and classifies in the image field, more high-dimensional information can be extracted along with the deepening of the neural network depth, so that the classification accuracy is improved, but the network degradation problems such as gradient explosion or gradient disappearance and the like can occur when the number of layers is too deep, so that a residual module needs to be added behind the network to solve the problems, and the loss function uses a cross entropy loss function:
Figure BDA0002889451500000122
wherein K is the number of species; y is a label, if the sample class is i, then yi1, otherwise equal to 0; p is a radical ofiIs the probability of class i being output by the classification model.
According to the medical image detection model training method, a priori medical image data set is obtained, a neural network is established according to the priori medical image data set, and a neural network segmentation model is trained; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the multi-scale convolution layer and residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of the medical care work.
The embodiment of the invention also provides a training device for the medical image detection model.
Fig. 7 is a block diagram of a medical image examination model training apparatus according to an embodiment, and as shown in fig. 6, the medical image examination model training apparatus according to an embodiment includes a module 100, a module 101, and a module 102:
a data set determining module 100 for obtaining a prior medical image to obtain a prior medical image data set;
the neural network establishing module 101 is used for establishing a neural network according to the prior medical image data set subjected to medical information classification and labeling so as to train a neural network segmentation model;
and the model determining module 102 is used for adding a multi-scale convolution layer and residual connection into the convolution network layer of the neural network segmentation model to obtain the medical image detection model.
The medical image detection model training device obtains a prior medical image data set, and establishes a neural network according to the prior medical image data set so as to train a neural network segmentation model; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the multi-scale convolution layer and residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of the medical care work.
The embodiment of the invention also provides a computer storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, the method for training the medical image detection model of any one of the embodiments is implemented.
Those skilled in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as a removable Memory device, a Random Access Memory (RAM), a Read-Only Memory (ROM), a magnetic disk, and an optical disk.
Alternatively, the integrated unit of the present invention may be stored in a computer-readable storage medium if it is implemented in the form of a software functional module and sold or used as a separate product. Based on such understanding, the technical solutions of the embodiments of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a terminal, or a network device) to execute all or part of the methods of the embodiments of the present invention. And the aforementioned storage medium includes: a removable storage device, a RAM, a ROM, a magnetic or optical disk, or various other media that can store program code.
Corresponding to the computer storage medium, in an embodiment, there is also provided a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the medical image detection model training methods in the embodiments.
The computer equipment obtains a prior medical image data set, and establishes a neural network according to the prior medical image data set so as to train a neural network segmentation model; and further, adding a multi-scale convolution layer and residual connection in a convolution network layer of the neural network segmentation model to obtain the medical image detection model. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the multi-scale convolution layer and residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of the medical care work.
The embodiment of the invention also provides a medical image detection method.
Fig. 8 is a flowchart illustrating a medical image detection method according to an embodiment, and as shown in fig. 8, a medical image processing method according to an embodiment includes steps S600 and S601:
s600, acquiring a medical image to be processed;
s601, inputting the medical image to be processed into a medical image detection model to obtain a medical image detection result.
It should be noted that, the processing modes of the medical image to be processed before and after being input into the medical image detection model are consistent with those of the prior medical image.
In the medical image detection method according to any of the embodiments, after the medical image to be processed is obtained, the medical image to be processed is input into the medical image detection model to obtain a medical image detection result, so that interference information is removed for diagnosis of medical staff. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
The embodiment of the invention provides a medical image detection device.
Fig. 9 is a block diagram of a medical image inspection apparatus according to an embodiment, and as shown in fig. 9, the medical image inspection apparatus according to an embodiment includes modules 200 and 201:
an image acquisition module 200, configured to acquire a medical image to be processed;
the image detection module 201 is configured to input the medical image to be processed into the medical image detection model to obtain a medical image detection result.
According to the medical image detection device, when the medical image to be processed is obtained, the medical image to be processed is input into the medical image detection model, so that a medical image detection result is obtained, and interference information is removed for diagnosis of medical staff. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
The embodiment of the present invention further provides another computer storage medium, on which computer instructions are stored, and the instructions, when executed by a processor, implement the medical image detection method according to any one of the above embodiments.
Those skilled in the art will understand that: all or part of the steps for implementing the method embodiments may be implemented by hardware related to program instructions, and the program may be stored in a computer readable storage medium, and when executed, the program performs the steps including the method embodiments; and the aforementioned storage medium includes: various media that can store program codes, such as a removable Memory device, a Random Access Memory (RAM), a Read-Only Memory (ROM), a magnetic disk, and an optical disk.
Alternatively, the integrated unit of the present invention may be stored in a computer-readable storage medium if it is implemented in the form of a software functional module and sold or used as a separate product. Based on such understanding, the technical solutions of the embodiments of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a terminal, or a network device) to execute all or part of the methods of the embodiments of the present invention. And the aforementioned storage medium includes: a removable storage device, a RAM, a ROM, a magnetic or optical disk, or various other media that can store program code.
Corresponding to the computer storage medium, in an embodiment, there is provided another computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the medical image detection methods in the embodiments.
According to the computer equipment, when the medical image to be processed is obtained, the medical image to be processed is input into the medical image detection model, so that a medical image detection result is obtained, and interference information is removed for diagnosis of medical staff. Based on the method, the medical image is segmented through the neural network segmentation model, the accuracy of image feature extraction is improved and the problem of network degradation is solved through the medical image detection model with the multi-scale convolution layer and the residual connection, medical information classification corresponding to the medical image is efficiently and accurately detected, and the burden of medical care work is reduced so as to be beneficial to improving the efficiency of medical care work.
The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.
The above examples only show some embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims (10)

1. A medical image detection model training method is characterized by comprising the following steps:
acquiring a prior medical image to obtain a prior medical image data set;
establishing a neural network according to the prior medical image data set subjected to medical information classification and labeling so as to train a neural network segmentation model;
and adding a multi-scale convolution layer and residual connection into the convolution network layer of the neural network segmentation model to obtain the medical image detection model.
2. The method for training a medical image detection model according to claim 1, further comprising, before the process of acquiring the prior medical image, the steps of:
and carrying out image preprocessing on the prior medical image.
3. The method for training a medical image detection model according to claim 1, wherein the process of obtaining a priori medical images to obtain a priori medical image data set comprises the steps of:
segmenting a prior region image of the prior medical image;
and obtaining a prior medical image data set according to the prior region image.
4. The method for training a medical image detection model according to claim 1, wherein the process of establishing a neural network according to the prior medical image data set after medical information classification and labeling comprises the steps of:
scaling the prior medical image in the prior medical image data set;
inputting the scaled prior medical image into a neural network, performing convolution processing on a convolution layer in the neural network, and calculating a characteristic subgraph;
and extracting features according to the feature subgraph through an interested region pooling layer in the neural network so as to carry out full-connection processing and classification processing of a logistic regression network.
5. The method of claim 4, wherein the logistic regression network has a loss function as follows:
Figure FDA0002889451490000021
wherein i represents anchors index, piRepresenting the forkround softmax mobility,
Figure FDA0002889451490000022
representing the corresponding GT prediction probability, i.e. IoU between the ith anchor and GT>0.7, the anchor is considered to be forego,
Figure FDA0002889451490000023
otherwise IoU<When 0.3, the anchor is considered to be background,
Figure FDA0002889451490000024
as for those 0.3<IoU<0.7 anchor does not participate in training; t represents a prediction bounding box,
Figure FDA00028894514900000210
represents a GT box corresponding to the forego anchor; whole LossThe method is divided into the following two parts:
the cls loss, namely the softmax loss calculated by rpn _ cls _ loss layer is used for network training for classifying anchors as forward and background;
the reg loss, namely soomth L1 loss calculated by rpn _ loss _ bbox layer, is used for bounding box regression network training; wherein, in reg loss, multiply
Figure FDA0002889451490000025
Shows regression with only forego anchors in mind.
6. The method of any one of claims 1 to 5, wherein the multi-scale convolution operation of the multi-scale convolution layer is as follows:
Figure FDA0002889451490000026
wherein the content of the first and second substances,
Figure FDA0002889451490000027
j-th output feature map of 1 x 1 convolution kernel connected with l-1 layers
Figure FDA0002889451490000028
And ith output feature map f of multi-scale convolution layeri 1*1，
Figure FDA0002889451490000029
For offset, n is the number of l-1 layer output profiles, ReLU () is the activation function, fi 3*3、fi 5*5The same is true.
7. The method according to claim 6, wherein the filling parameter of the 1 x 1 convolution kernel is 0, the filling parameter of the 3 x 3 convolution kernel is 1, and the filling parameter of the 5 x 5 convolution kernel is 2;
the output characteristic diagram of the multi-scale convolutional layer is as follows:
fi Multi_Scale＝Concat(fi 1*1，fi 3*3，fi 5*5)。
8. the method of any one of claims 1 to 5, wherein the loss function corresponding to the residual connection is a cross-entropy loss function, and is as follows:
Figure FDA0002889451490000031
wherein K is the number of species; y is a label, if the sample class is i, then yi1, otherwise equal to 0; p is a radical ofiIs the probability of class i being output by the classification model.
9. A medical image detection model training device is characterized by comprising:
the data set determining module is used for acquiring a prior medical image to obtain a prior medical image data set;
the neural network establishing module is used for establishing a neural network according to the prior medical image data set subjected to medical information classification and labeling so as to train a neural network segmentation model;
and the model determining module is used for adding a multi-scale convolution layer and residual connection into the convolution network layer of the neural network segmentation model to obtain the medical image detection model.
10. A medical image detection method is characterized by comprising the following steps:
acquiring a medical image to be processed;
and inputting the medical image to be processed into a medical image detection model to obtain a medical image detection result.
CN202110023647.6A 2021-01-08 2021-01-08 Medical image detection model training method, medical image detection method and device Pending CN112766332A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110023647.6A CN112766332A (en) 2021-01-08 2021-01-08 Medical image detection model training method, medical image detection method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110023647.6A CN112766332A (en) 2021-01-08 2021-01-08 Medical image detection model training method, medical image detection method and device

Publications (1)

Publication Number Publication Date
CN112766332A true CN112766332A (en) 2021-05-07

Family

ID=75700987

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110023647.6A Pending CN112766332A (en) 2021-01-08 2021-01-08 Medical image detection model training method, medical image detection method and device

Country Status (1)

Country Link
CN (1) CN112766332A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115222688A (en) * 2022-07-12 2022-10-21 广东技术师范大学 Medical image classification method based on graph network time sequence

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115222688A (en) * 2022-07-12 2022-10-21 广东技术师范大学 Medical image classification method based on graph network time sequence
CN115222688B (en) * 2022-07-12 2023-01-10 广东技术师范大学 Medical image classification method based on graph network time sequence

Similar Documents

Publication Publication Date Title
CN108648172B (en) CT (computed tomography) map pulmonary nodule detection system based on 3D-Unet
CN111325739B (en) Method and device for detecting lung focus and training method of image detection model
CN110706246B (en) Blood vessel image segmentation method and device, electronic equipment and storage medium
Murakami et al. Automatic identification of bone erosions in rheumatoid arthritis from hand radiographs based on deep convolutional neural network
US10853409B2 (en) Systems and methods for image search
CN111640120A (en) Pancreas CT automatic segmentation method based on significance dense connection expansion convolution network
CN113012173A (en) Heart segmentation model and pathology classification model training, heart segmentation and pathology classification method and device based on cardiac MRI
CN113744183B (en) Pulmonary nodule detection method and system
RU2654199C1 (en) Segmentation of human tissues in computer image
US20230005140A1 (en) Automated detection of tumors based on image processing
KR20200082660A (en) Pathological diagnosis method and apparatus based on machine learning
CN110570394A (en) medical image segmentation method, device, equipment and storage medium
CN113192067B (en) Intelligent prediction method, device, equipment and medium based on image detection
CN112767403A (en) Medical image segmentation model training method, medical image segmentation method and device
CN112766332A (en) Medical image detection model training method, medical image detection method and device
CN112750110A (en) Evaluation system for evaluating lung lesion based on neural network and related products
Delmoral et al. Segmentation of pathological liver tissue with dilated fully convolutional networks: A preliminary study
CN114627136B (en) Tongue image segmentation and alignment method based on feature pyramid network
CN112862786B (en) CTA image data processing method, device and storage medium
CN114359308A (en) Aortic dissection method based on edge response and nonlinear loss
CN110570417B (en) Pulmonary nodule classification device and image processing equipment
Al-Eiadeh Automatic lung field segmentation using robust deep learning criteria
CN112862785A (en) CTA image data identification method, device and storage medium
CN112766333B (en) Medical image processing model training method, medical image processing method and device
CN116777893B (en) Segmentation and identification method based on characteristic nodules of breast ultrasound transverse and longitudinal sections

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination