CN115762721A - Medical image quality control method and system based on computer vision technology - Google Patents
Medical image quality control method and system based on computer vision technology Download PDFInfo
- Publication number
- CN115762721A CN115762721A CN202211318143.8A CN202211318143A CN115762721A CN 115762721 A CN115762721 A CN 115762721A CN 202211318143 A CN202211318143 A CN 202211318143A CN 115762721 A CN115762721 A CN 115762721A
- Authority
- CN
- China
- Prior art keywords
- image
- quality control
- medical image
- model
- 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.)
- Pending
Links
Images
Landscapes
- Medical Treatment And Welfare Office Work (AREA)
Abstract
The invention provides a medical image quality control method and system based on a computer vision technology, which comprises the following steps: collecting a plurality of historical image inspection data of a medical institution within a medical image mutual-recognition area range within a preset time period; extracting DR images in the historical image inspection data, preprocessing the DR images, and generating a marked DR image set with the same angle; constructing an initial DR image recognition model for quality control evaluation by using a computer vision technology, and training, verifying and testing the initial DR image recognition model based on a DR image set to generate a DR image prediction module; acquiring a DR image to be evaluated based on medical image data to be evaluated, and inputting the DR image to be evaluated into a DR image prediction module to acquire a prediction result; the prediction result comprises various indexes of image quality control; and scoring each index based on a preset medical image quality control evaluation rule, so as to output a quality control evaluation result of the DR image to be evaluated. The invention can greatly improve the quality control evaluation efficiency of the medical image.
Description
Technical Field
The invention relates to the field of medical image quality control, in particular to a medical image quality control method and system based on a computer vision technology.
Background
The quality of medical images directly affects the diagnosis and treatment decision of medical workers on diseases, and the homogenization of medical quality is the basis of the sharing of examination data and the mutual recognition of examination results. In the traditional medical image quality control task, the quality of medical image data is uneven, huge manpower is consumed for quality control work, and the evaluation result depends on the experience of experts seriously, so that the standardization degree of the evaluation result is low, and the full-scale evaluation is difficult to realize, thereby hindering the implementation of the image quality control work.
In recent years, deep learning models for DR image quality control are researched in the fields of image classification and image semantic segmentation, but in the current quality control methods based on deep learning, methods are designed only for individual typical quality index evaluation tasks, the evaluation methods are single, all quality control index requirements cannot be covered, and the method is difficult to be applied to diversified medical image quality control tasks.
Disclosure of Invention
In view of the above, the present invention provides a medical image quality control method and system based on computer vision technology to improve the above problems.
In one aspect, the present invention provides a medical image quality control method based on computer vision technology, wherein the medical image comprises a DR image, and the method comprises:
collecting a plurality of historical image inspection data of a medical institution within a medical image mutual-recognition area range within a preset time period;
extracting DR images in the historical image inspection data, preprocessing the DR images, and generating a marked DR image set with the same angle;
constructing an initial DR image recognition model for quality control evaluation by using a computer vision technology, and training, verifying and testing the initial DR image recognition model based on a DR image set to generate a DR image prediction module;
acquiring a DR image to be evaluated based on medical image data to be evaluated, and inputting the DR image to be evaluated into a DR image prediction module to acquire a prediction result; the prediction result comprises various indexes of image quality control; and scoring each index based on a preset medical image quality control evaluation rule, so as to output a quality control evaluation result of the DR image to be evaluated.
Further, the method for preprocessing the DR image comprises the following steps:
presetting a medical image quality control standard;
constructing a medical image labeling semantic description frame based on the medical image quality control standard; the medical image labeling semantic description frame at least comprises a body position specification, an image layout and an integral image quality specification;
based on the medical image labeling semantic description framework, performing quality control label labeling on the DR image set to obtain a labeled DR image set, wherein the quality control label at least comprises classification labeling, region sketching labeling and position information labeling; the marked DR image comprises DR image sample data, a quality control label and a corresponding relation of the DR image sample data and the quality control label.
Further, preprocessing the DR image further includes:
after each DR image in the marked DR image set is filled to a uniform length, normalization is carried out; and aligning each DR image to a uniform angle through an affine transformation algorithm so as to generate a marked DR image set with the same angle.
Further, the initial DR image identification model comprises a classification model for predicting the artifacts of the DR image and the overall image quality; the method for constructing the classification model comprises the following steps:
constructing a deep convolutional neural network of a classification model taking a convolutional layer, a pooling layer, a normalization layer, an activation function and a full-link layer as a main network; a classification activation function is connected behind the full connection layer; the output dimension of the full connection layer is consistent with the quantity of the quality control labels; the loss function used by the classification model is a logistic regression loss function or an L2 loss function;
and pre-training the initial DR image recognition model by taking the open source picture library as a training set to obtain the initialized network parameters of the classification model.
Further, the deep convolutional neural network of the classification model is AlexNet or ResNet or DenseNet.
Further, the initial DR image recognition model further comprises a key point detection model, wherein the key point detection model is used for predicting key point coordinates with specific semantics in the DR image; the construction method of the key point detection model comprises the following steps:
constructing a deep neural network of a key point detection model based on a coding and decoding network U-Net as a basic network architecture; the deep neural network comprises a convolution layer, a pooling layer, a normalization layer, an activation function and an up-sampling layer;
adding an attention layer between an up-sampling module and a corresponding down-sampling module of a decoder of the coding and decoding network; respectively adding a 3 x 3 convolution and nonlinear activation unit after an input module and an output module of a coding and decoding network to obtain a key point detection model; the loss function used by the key point detection model is a cross entropy loss function or an L2 loss function.
Further, constructing an initial DR image recognition model for quality control evaluation further comprises a segmentation model for segmenting an image region related to the medical image diseases in the medical image; the segmentation model construction method comprises the following steps:
constructing a deep neural network of a segmentation model based on a basic network architecture of a coding and decoding network U-Net; the method comprises the steps that a deep neural network trunk network is initialized, wherein the deep neural network trunk network comprises a convolution layer, a pooling layer, a normalization layer, an activation function and an up-sampling layer; the loss function used by the segmentation model is a cross-entropy loss function or an L2 loss function.
Further, the preset time period is set according to the mutual recognition timeliness specified by the health and health management department in the medical image mutual recognition area.
Further, based on a preset medical image quality control evaluation rule, scoring is performed on each index, so that a quality control evaluation result of the DR image to be evaluated is output, and the method specifically comprises the following steps:
according to the target part corresponding to the DR image, a plurality of quality control tasks are preset, and each quality control task comprises: quality control standards, involved parts and judgment conditions;
and calculating the score value of each quality control standard corresponding to the DR image to be evaluated according to the prediction result of the DR image to be evaluated by the DR image prediction module and the judgment condition of the quality control task, and obtaining the quality control evaluation result of the DR image to be evaluated through weighted calculation.
Further, constructing a basic information integrity evaluation model;
extracting basic information of an examination item of the medical image data to be evaluated, wherein the basic information at least comprises the following components: basic information, medical history, examination part, examination method, examination item name and unique hospital identification code of the examined person;
predicting the basic information of the examination item of the medical image data to be evaluated based on the basic information integrity evaluation model to obtain a prediction result; and scoring the prediction result according to the preset medical image quality control evaluation rule to obtain a basic information integrity scoring result.
And further, combining the basic information integrity evaluation result and the DR image quality control evaluation result to automatically generate a medical image quality control evaluation report corresponding to the medical image data to be evaluated.
In another aspect, a medical image quality control system based on computer vision technology is provided, the medical image includes a DR image, including:
the acquisition module is used for acquiring a plurality of historical image inspection data of the medical institution within the medical image mutual-recognition area range within a preset time period;
a pretreatment module: the preprocessing module is used for extracting DR images in the historical image inspection data, preprocessing the DR images and generating marked DR image sets with the same angle;
a model generation module: the model generation module utilizes a computer vision technology to construct an initial DR image recognition model for quality control evaluation, and trains, verifies and tests the initial DR image recognition model based on a DR image set to generate a DR image prediction module;
DR image quality control module: the quality control module acquires a DR image to be evaluated based on the medical image data to be evaluated, inputs the DR image to be evaluated into the DR image prediction module, and acquires a prediction result; the prediction result comprises various indexes of image quality control; and scoring each index based on a preset medical image quality control evaluation rule, so as to output a quality control evaluation result of the DR image to be evaluated.
Further, the method also comprises the following steps: a basic information integrity quality control module;
the basic information integrity quality control module is used for constructing a basic information integrity evaluation model;
extracting basic information of an inspection item of medical image data to be evaluated, wherein the basic information at least comprises the following components: basic information, medical history, examination part, examination method, examination item name and unique hospital identification code of the examined person;
predicting the basic information of the examination item of the medical image data to be evaluated based on the basic information integrity evaluation model to obtain a prediction result; and scoring the prediction result according to a preset medical image quality control evaluation rule to obtain a basic information integrity scoring result.
The medical image quality control report generation module is used for automatically generating a medical image quality control evaluation report corresponding to the medical image data to be evaluated in combination with the evaluation result generated by the basic information integrity quality control module and the evaluation result generated by the DR image quality control module.
The invention has the following beneficial effects:
the invention utilizes the computer vision neural network to assist in realizing the standardized and intelligent classification, identification and grading of DR images; the medical image quality control evaluation rule based on the standard medical image inspection mutual recognition system evaluates the quality control indexes including but not limited to those mentioned herein, thereby ensuring the standard property of the medical image evaluation and improving the evaluation accuracy and the evaluation efficiency. Based on this, the method solves the problem that the identification and evaluation are carried out through a large amount of manpower in the prior art. The method can assist quality control experts in finding problems, help accurately, improve mutual recognition rate and achieve consistency and effectiveness of inspection result mutual recognition data sharing.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments will be briefly described below, it should be understood that the following drawings only illustrate some embodiments of the present invention and therefore should not be considered as limiting the scope, and for those skilled in the art, other related drawings can be obtained according to the drawings without inventive efforts.
Fig. 1 is a system flowchart of a medical image quality control method based on computer vision technology according to a first embodiment of the present invention.
Fig. 2 is a flowchart illustrating a medical image examination data normalization process based on computer vision technology according to a first embodiment of the present invention.
Fig. 3 is a schematic diagram of the serial numbers and semantic labeling results of the key points in the chest orthostatic DR image according to the first embodiment of the present invention.
Fig. 4 is a schematic diagram of a classification model network structure, which is provided by a ResNet network according to a first embodiment of the present invention.
Fig. 5 is a schematic diagram of a network structure of a keypoint detection model according to a first embodiment of the present invention.
Fig. 6 is a lung segmentation diagram of a chest DR image according to a first embodiment of the present invention.
Fig. 7 is a schematic diagram of a segmentation model network structure according to a first embodiment of the present invention.
Fig. 8 is a schematic structural diagram of a system according to a second embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to better understand the technical scheme of the invention, the following detailed description of the embodiments of the invention is made with reference to the accompanying drawings.
It should be understood that the described embodiments are only some embodiments of the invention, and not all embodiments. All other embodiments, which can be obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
The terminology used in the embodiments of the invention is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. As used in the examples of the present invention and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
The first embodiment of the invention utilizes three tasks and visual technology in the field of computer vision to assist in realizing standardized and intelligent classification, identification and grading of DR images. The quality control evaluation rules of the medical image based on the standard medical image inspection mutual recognition system are used for evaluating the quality control indexes including but not limited to the quality control indexes mentioned herein, so that the standard property of medical image evaluation is ensured, and the evaluation accuracy and the evaluation efficiency are improved. The three major tasks in the field of computer vision include: 1. image classification, that is, information for structuring an image into a certain category, describes a picture with a previously determined category or instance ID. 2. Object detection, which focuses on a specific object, requires that both the category information and the location information of this object be obtained. Compared with the classification task, detection gives understanding of the foreground and the background of the picture, and an object of important interest needs to be separated from the background, and description (category and position) of the object is determined, so that the output of a detection model is a list, and each item of the list gives the category and position of the detected object by using an array. 3. And (5) image segmentation. The segmentation comprises semantic segmentation and instance segmentation, is the expansion of a target detection task and requires to describe the outline of a target.
The invention is described in further detail below with reference to the following detailed description and accompanying drawings: the quality control method according to the present invention can perform quality control evaluation on DR images of a plurality of parts of a human body, and the following description will be given by taking only a chest normal DR image as an example.
Referring to fig. 1, a medical image quality control method based on computer vision technology according to a first embodiment of the present invention is described with reference to a DR image, and the method includes:
SP1: collecting a plurality of historical image inspection data of a medical institution within a medical image mutual-recognition area range within a preset time period; the preset time period is set according to the mutual recognition timeliness specified by the health management department in the medical image mutual recognition area.
Specifically, the historical image inspection data sources here are: historical image inspection data of the medical institution within a preset coverage range and a preset time period are collected through a special health network and stored on a regional medical data cloud platform. The preset coverage area may be a part of medical institutions in the area (such as three-level medical institutions in the area), or may be all medical institutions in the area, and is specifically determined according to a clear mutual recognition institution range of the health and health management department in the area. The preset time period can also be adjusted according to the clear mutual recognition aging of the regional health management department. The historical image examination data includes, but is not limited to, examination application forms, examination reports and image files, and the examination application forms, examination reports and image files corresponding to a single examination of the examinee are taken as a piece of historical examination data.
SP2: constructing an image quality control labeling semantic description frame to obtain an image quality control standard;
based on the existing image mutual-recognition quality control standard, a medical image quality control labeling semantic description framework is constructed for the chest orthostatic quality control task through an authoritative medical image expert team. As shown in table 1; the quality control labeling semantic description information mainly comprises: the integrity of the basic image information, the body position specification, the image layout and the overall image quality specification. Each quality control index contained in the image quality control labeling semantic description frame is an image quality index which needs to be paid attention by a technician in the shooting process, and the medical image follows the image quality control standard, which is the premise of realizing image inspection mutual recognition or accurate diagnosis by a doctor.
Table 1 medical image quality control annotation semantic description framework takes Chongqing regional image inspection mutual recognition standard as an example:
SP3: medical image examination data is standardized. Doctors in different hospitals have different default writing specifications and writing habits, and different hospitals have different management specifications and informatization degrees, so that the examination information contained in the same examination item in different hospitals has larger difference. Therefore, it is necessary to use methods such as statistics, artificial feature engineering, machine learning, and natural language processing to construct a mapping from non-standard medical examination item information to standard medical examination mutual-recognition item information, and the standardized flow chart is shown in fig. 2.
SP4: evaluating the integrity of basic information of the examination item basic information of the medical image data to be evaluated; the examination item basic information includes at least subject basic information and examination basic information.
S41, constructing a basic information integrity evaluation model;
and S42, extracting basic information of the examination item of the medical image data to be evaluated, wherein relevant fields of the basic information to be evaluated in the examination item information comprise but are not limited to an examination part, an examination method, an examination item name, an examination item code, an examination item charging code, a medical image examination description, a medical image examination condition specific description, an examination result, a medical history, an examination equipment corresponding code, a unique hospital identification code and the like.
S43: and predicting the basic information of the examination item of the medical image data to be evaluated based on the basic information integrity evaluation model to obtain a prediction result. And scoring the prediction result according to a preset medical image quality control evaluation rule so as to generate a basic information integrity evaluation result.
SP5: performing quality control evaluation on a DR image in medical image data to be evaluated;
s51: extracting DR images in the historical image inspection data, preprocessing the DR images, and generating a marked DR image set with the same angle;
wherein, carry out the preliminary treatment to DR image, include: presetting a medical image quality control standard; constructing a medical image labeling semantic description frame based on the medical image quality control standard; the medical image labeling semantic description frame at least comprises a body position specification, an image layout and an integral image quality specification; based on the medical image labeling semantic description framework, performing quality control label labeling on the DR image set to obtain a labeled DR image set, wherein the quality control label at least comprises classification labeling, region sketching labeling and position information labeling; the marked DR image comprises DR image sample data, a quality control label and a corresponding relation of the DR image sample data and the quality control label. In addition, the method for preprocessing the DR image further comprises the following steps: after each DR image in the marked DR image set is filled to a uniform length, normalization is carried out; and aligning each DR image to a uniform angle through an affine transformation algorithm so as to generate a marked DR image set with the same angle.
Specifically, taking a chest normal DR image as an example, the Dicom file is analyzed to obtain a normal DR chest image, all DR image data are supplemented to a uniform length for normalization, and then all DR images are aligned to a uniform angle through affine transformation. The so-called affine transformation can be realized by a complex of a series of atomic transformations, including: translation (Translation), scaling (Scale), flip (Flip), rotation (Rotation), and shearing (Shear). The acquired DR chest image data comprises DR image sample data, a quality control label (see table 2) assessed by an expert and a corresponding relation of the DR chest image sample data and the quality control label. The DR image set was classified into a training set, a validation set, and a test set according to 8.
And S52, classifying the DR image set, performing semantic segmentation and labeling key points.
And (4) carrying out classification labeling, region sketching labeling and position information labeling on the DR image data by using a labeling tool by experts strictly surrounding the established quality control standard. Taking the position information of the chest DR image as an example, the labeling effect, the key point sequence number and the semantic corresponding table 2 are shown in FIG. 3;
TABLE 2 Key points sequence number and semantic correspondence Table
S53: constructing an initial DR image recognition model for quality control evaluation by using a computer vision technology, and training, verifying and testing the initial DR image recognition model based on a DR image set to generate a DR image prediction module; the initial DR image identification model comprises a classification model used for predicting the artifacts of the DR image and the overall image quality; the key point detection model is used for predicting key point coordinates with specific semantics in the DR image; and a segmentation model for segmenting an image region in the medical image that is associated with the medical image condition.
S531, a classification model construction method, please refer to FIG. 4;
constructing a deep convolutional neural network of a classification model taking a convolutional layer, a pooling layer, a normalization layer, an activation function and a full-link layer as a main network; a classification activation function is connected behind the full connection layer; the output dimension of the full connection layer is consistent with the quantity of the quality control labels; the loss function used by the classification model is a logistic regression loss function or an L2 loss function; and pre-training the initial DR image recognition model by taking the open source picture library as a training set to obtain the initialized network parameters of the classification model. The deep convolution neural network of the classification model is AlexNet or ResNet or DenseNet.
Specifically, the logistic regression loss function is taken as an example for explanation, and the formula is shown as follows:
wherein X is the input preprocessed image data, (y) 1 ,y 2 ,...,y k ) Labeling the result for the expert corresponding to the X,
and predicting the probability (belonging to a classification task) or the quality control score (belonging to a regression task) whether the X meets the quality control standard of the ith type or not for the deep neural model.
The main network of the convolutional neural network consists of a convolutional layer, a pooling layer, a normalization layer, a full connection layer FC, an activation function ReLU and Softmax. The final two layers of the convolutional neural network, including the pooling layer and the fully-connected layer, output results of which are further used for feature selection and mapping. Available deep convolutional neural network models include AlexNet, resNet, denseNet and the like, and the selection of a specific model can be flexibly selected according to conditions such as the data volume of medical images and available computing resources. Optionally, the network parameters may be initialized via pre-training. For example, model initialization is performed with model weights pre-trained on the ImageNet dataset to take full advantage of the ability to extract deep features learned on the ImageNet dataset.
S532, constructing a key point detection model, please refer to FIG. 5;
constructing a deep neural network of a key point detection model based on a basic network architecture of a coding and decoding network U-Net; the deep neural network comprises a convolution layer, a pooling layer, a normalization layer, an activation function and an up-sampling layer; adding an attention layer between an up-sampling module and a corresponding down-sampling module of a decoder of the coding and decoding network; respectively adding a 3 x 3 convolution and nonlinear activation unit after an input module and an output module of a coding and decoding network to obtain a key point detection model; the loss function used by the key point detection model is a cross entropy loss function or an L2 loss function.
Specifically, the task of this type accurately predicts the coordinates of key points with specific semantics in the chest DR image in the image through a deep learning technique, such as the left clavicle end, the left lung apex, the right costal diaphragm angle, and the like. Constructing a deep neural network, taking a codec network U-Net as a basic network architecture, and modifying the deep neural network as follows:
(1) Adding an attention module between each decoder upsampling module and the corresponding downsampling module, as shown in fig. 4;
(2) Add a 3 x 3 convolution and nonlinear activation unit ReLU module to each of the first module (input) and the last module (output) of U-Net. The input of the deep neural network model is an image after preprocessing, and the output is a thermodynamic diagram. The main network of the key point detection model consists of a convolution layer, a pooling layer, a normalization layer, an activation function and an up-sampling layer, wherein a Sigmoid activation function is added to an output layer of the network, and the output of the model is mapped to a range from 0 to 1.
The loss function used by the key point detection model can be a cross entropy loss function, an L2 loss function and the like, and the first embodiment of the invention does not limit the specific used loss function type and can be flexibly selected according to the actual application scene requirement and the final model effect.
Taking the cross entropy loss function as an example, the formula is as follows:
wherein the content of the first and second substances,
labeling the result of the expert at the image position (i, j),
is the prediction result of the deep neural model at image position (i, j). The selection of the number of channels of each layer of the specific model can be flexibly selected according to the conditions of the data volume of the medical image, available computing resources and the like. Optionally, the network parameters may be initialized via pre-training. For example, model initialization is performed with model weights pre-trained on the ImageNet dataset to take full advantage of the ability to extract deep features learned on the ImageNet dataset.
S533, a segmentation model construction method, please refer to FIG. 7;
this type of task segments a Region of Interest (ROI) of a technician or doctor from a DR image by a deep learning technique. Similar to the construction method of the key point detection model, the method is based on the U-Net network structure, but the difference is mainly in loss functions of output and training.
Constructing a deep neural network of a segmentation model based on a basic network architecture of a coding and decoding network U-Net; the method comprises the steps that a deep neural network trunk network is initialized, wherein the deep neural network trunk network comprises a convolution layer, a pooling layer, a normalization layer, an activation function and an up-sampling layer; the loss function used by the segmentation model is a cross-entropy loss function or an L2 loss function.
Specifically, taking the chest DR image segmentation as an example, fig. 6 is a schematic diagram of lung segmentation. And (3) constructing a deep neural network, wherein the segmentation model takes U-Net as a network architecture, as shown in figure 6. The model input size is 512x512, and the output size is consistent with the input size. The loss function used by the segmentation model may be a cross entropy loss function, an L2 loss function, or the like, and the first embodiment of the present invention does not limit the type of the specifically used loss function, and can be flexibly selected according to the actual application scenario requirements and the final model effect. The selection of the number of channels of each layer of the specific model can be flexibly selected according to the conditions of the data volume of the medical image, available computing resources and the like. Optionally, the network parameters may be initialized via pre-training. For example, model initialization is performed with model weights pre-trained on the ImageNet dataset to take full advantage of the ability to extract deep features learned on the ImageNet dataset.
S54, training and verifying the classification model, the key point detection model and the segmentation model respectively; when the model is trained, the batch processing quantity and the initial learning rate are set. Optionally, the adjustment strategy for the model training learning rate may select manual adjustment or strategy adjustment, where the strategy adjustment includes learning rate attenuation adjustment (such as Step adjustment at equal intervals) of a fixed strategy and adaptive learning rate attenuation adjustment (such as ReduceLRonPlateau). The selection may be based on model convergence speed versus performance variation. The learning rate attenuation adjustment of the fixed strategy is taken as an example for explanation: and when the loss function of the verification set does not fall for two continuous periods, reducing the learning rate to be one nth of the original learning rate, and ending the training until the loss function does not fall for five continuous periods. The best model parameters are saved for each cycle of model training. And (5) verifying the trained deep neural network model through the verification set obtained in the steps S51-S52, and finishing training when the loss function of the verification set reaches a preset convergence condition, so that the DR image prediction module is packaged and is on line.
S55, acquiring a DR image to be evaluated based on the medical image data to be evaluated; inputting a DR image to be evaluated into the DR image prediction module to obtain a prediction result; the prediction result comprises various indexes of medical image quality control; and scoring each quality control index based on a preset medical image quality control evaluation rule, so as to output a quality control evaluation result of the DR image to be evaluated.
Wherein, based on preset medical image quality control evaluation rule, score each quality control index to output the quality control evaluation result of the DR image that waits to evaluate, specifically include: according to the target part corresponding to the DR image, a plurality of quality control tasks are preset, wherein each quality control task comprises: quality control indexes, related parts and judgment conditions. And calculating the score value of each quality control index corresponding to the DR image to be evaluated according to the prediction result of the DR image to be evaluated by the DR image prediction module and the judgment condition of the quality control task, and obtaining the quality control evaluation result of the DR image to be evaluated through weighted calculation.
Performing quality control characteristic classification on the DR chest position image by using a classification model of a DR image prediction module; and acquiring image characteristics by using the deep neural network so as to judge the image quality of the corresponding quality control task. And after the coordinates and the segmentation result of each key point on the image are obtained by using the deep neural network, judging the image quality of each quality control task according to the coordinates of the key points, the segmentation result and the corresponding judgment condition.
Taking the chest DR image quality control as an example, the image quality corresponding to each quality control task can be determined according to the chest key point coordinates, the lung segmentation result and the corresponding determination conditions. The different quality control tasks, the involved parts and the determination conditions are shown in table 3.
TABLE 3 quality control tasks and related keypoints and decision rules
Wherein p is i Model pair i th The coordinates of the point prediction are taken into account,
is i th The x-coordinate of the point or points,
is i th The y coordinate of the point, H is the image height, W is the image width, δ is the physical distance (in centimeters) corresponding to the unit pixel, and τ is the threshold value for determination, which can be set by the user according to experience or obtained by a ten-fold intersection method (note that, for each quality control standard, the threshold value τ is not necessarily the same, and the same symbol is used in the table only for convenience of writing).
And SP6, combining the evaluation result of the integrity of the basic information and the quality control evaluation result of the DR image, and automatically generating a medical image quality control evaluation report corresponding to the medical image data to be evaluated.
Specifically, the medical image quality control evaluation rule is set for the score of the quality control semantic description frame of each target quality control evaluation item. And combining the medical image quality evaluation result obtained in the step SP5 and the medical image information integrity evaluation result obtained in the step SP4 to automatically generate a medical image quality control evaluation report corresponding to each piece of medical image data to be evaluated.
In conclusion, the invention utilizes the computer vision neural network to assist in realizing the standardized and intelligent classification, identification and grading of DR images; the medical image quality control evaluation rule based on the standard medical image inspection mutual recognition system evaluates the quality control indexes including but not limited to those mentioned herein, thereby ensuring the standard property of the medical image evaluation and improving the evaluation accuracy and the evaluation efficiency. Based on this, the method solves the problem that the identification and evaluation are carried out through a large amount of manpower in the prior art. The method can assist quality control experts in finding problems, accurately help, improve mutual recognition rate, realize consistency and effectiveness of inspection result mutual recognition data sharing, improve mutual recognition rate, and promote implementation of inspection data sharing and inspection result mutual recognition.
Referring to fig. 8, a second embodiment of the present invention provides a medical image quality control system based on computer vision technology, wherein the medical image includes a DR image, including:
the acquisition module 110: the acquisition module is used for acquiring a plurality of historical image inspection data of the medical institution within the medical image mutual-recognition area range within a preset time period;
the preprocessing module 120: the preprocessing module is used for extracting DR images in the historical image inspection data, preprocessing the DR images and generating marked DR image sets with the same angle;
model generation module 130: the model generation module utilizes a computer vision technology to construct an initial DR image recognition model for quality control evaluation, and trains, verifies and tests the initial DR image recognition model based on a DR image set to generate a DR image prediction module;
the DR image quality control module 140: the quality control module acquires a DR image to be evaluated based on the medical image data to be evaluated, inputs the DR image to be evaluated into the DR image prediction module, and acquires a prediction result; the prediction result comprises various indexes of image quality control; and scoring each index based on a preset medical image quality control evaluation rule, so as to output a quality control evaluation result of the DR image to be evaluated.
In addition, the method further comprises the following steps: a basic information integrity quality control module;
the model building module 150: the model construction module is used for constructing a basic information integrity evaluation model;
the information extraction module 160: the information extraction module is used for extracting basic information of the examination item of the medical image data to be evaluated, and the basic information at least comprises: basic information, medical history, examination part, examination method, examination item name and unique hospital identification code of the examined person;
the basic information integrity quality control module 170: the basic information integrity quality control module is used for predicting the basic information of the examination item of the medical image data to be evaluated based on the basic information integrity evaluation model to obtain a prediction result; and scoring the prediction result according to the preset medical image quality control evaluation rule to obtain a basic information integrity scoring result.
In addition, the medical image quality control report generation module 180 is further included, and the medical image quality control report generation module is configured to automatically generate a medical image quality control evaluation report corresponding to the medical image data to be evaluated in combination with the evaluation result generated by the basic information integrity degree quality control module and the evaluation result generated by the DR image quality control module.
In the embodiments provided in the embodiments of the present invention, it should be understood that the disclosed apparatus and method may be implemented in other manners. The apparatus and method embodiments described above are illustrative only, as the flowcharts and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.
In addition, the functional modules in the embodiments of the present invention may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
This function, if implemented in the form of a software function module and sold or used as a separate product, may be stored in a computer-readable storage medium. Based on such understanding, the technical solution 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, an electronic device, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes. It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (14)
1. A medical image quality control method based on computer vision technology, the medical image comprises a DR image, and the method is characterized by comprising the following steps:
collecting a plurality of historical image inspection data of a medical institution within a medical image mutual-recognition area range within a preset time period;
extracting DR images in the historical image inspection data, preprocessing the DR images, and generating a marked DR image set with the same angle;
constructing an initial DR image recognition model for quality control evaluation by using a computer vision technology, and training, verifying and testing the initial DR image recognition model based on the DR image set to generate a DR image prediction module;
acquiring a DR image to be evaluated based on medical image data to be evaluated, and inputting the DR image to be evaluated into the DR image prediction module to acquire a prediction result; the prediction result comprises various indexes of image quality control; and scoring each index based on a preset medical image quality control evaluation rule, so as to output a quality control evaluation result of the DR image to be evaluated.
2. The medical image quality control method based on computer vision technology as claimed in claim 1, wherein the preprocessing of the DR image comprises:
presetting a medical image quality control standard;
constructing a medical image labeling semantic description frame based on the medical image quality control standard; the medical image labeling semantic description frame at least comprises a body position specification, an image layout and an overall image quality specification;
based on the medical image labeling semantic description framework, performing quality control label labeling on the DR image set to obtain a labeled DR image set, wherein the quality control label at least comprises classification labeling, region sketching labeling and position information labeling; the marked DR image comprises DR image sample data, a quality control label and a corresponding relation of the DR image sample data and the quality control label.
3. The medical image quality control method based on computer vision technology as claimed in claim 2, wherein the DR image is preprocessed, further comprising:
after each DR image in the marked DR image set is filled to a uniform length, normalization is carried out; and aligning each DR image to a uniform angle through an affine transformation algorithm to generate the marked DR image set with the same angle.
4. The medical image quality control method based on the computer vision technology as claimed in claim 1, wherein: the initial DR image identification model comprises a classification model used for predicting the artifacts of the DR image and the overall image quality; the classification model construction method comprises the following steps:
constructing a deep convolutional neural network of a classification model taking a convolutional layer, a pooling layer, a normalization layer, an activation function and a full-link layer as a main network; a classification activation function follows the full connection layer; the output dimension of the full connection layer is consistent with the quantity of the quality control labels; the loss function used by the classification model is a logistic regression loss function or an L2 loss function;
and pre-training the initial DR image recognition model by taking an open source picture library as a training set to obtain the initial network parameters of the classification model.
5. The medical image quality control method based on the computer vision technology as claimed in claim 4, wherein: the deep convolution neural network of the classification model is AlexNet or ResNet or DenseNet.
6. The medical image quality control method based on the computer vision technology as claimed in claim 1, wherein: the initial DR image recognition model further comprises a key point detection model, and the key point detection model is used for predicting key point coordinates with specific semantics in the DR image; the construction method of the key point detection model comprises the following steps:
constructing a deep neural network of a key point detection model based on a basic network architecture of a coding and decoding network U-Net; the main network of the deep neural network comprises a convolution layer, a pooling layer, a normalization layer, an activation function and an up-sampling layer;
adding an attention layer between an up-sampling module and a corresponding down-sampling module of a decoder of the codec network; respectively adding a 3 x 3 convolution and nonlinear activation unit behind an input module and an output module of the coding and decoding network to obtain the key point detection model; the loss function used by the key point detection model is a cross entropy loss function or an L2 loss function.
7. The medical image quality control method based on the computer vision technology as claimed in claim 1, wherein: the method comprises the following steps that an initial DR image recognition model for quality control evaluation is constructed, and the initial DR image recognition model further comprises a segmentation model for segmenting an image area related to medical image symptoms in a medical image; the segmentation model construction method comprises the following steps:
constructing a deep neural network of a segmentation model based on a basic network architecture of a coding and decoding network U-Net; the deep neural network comprises a convolution layer, a pooling layer, a normalization layer, an activation function and an up-sampling layer; the loss function used by the segmentation model is a cross entropy loss function or an L2 loss function.
8. The medical image quality control method based on the computer vision technology as claimed in claim 1, wherein: and the preset time period is set according to the mutual recognition timeliness specified by the health management department of the medical image mutual recognition area.
9. The medical image quality control method based on the computer vision technology as claimed in claim 1, wherein: the method includes the steps of scoring each index based on a preset medical image quality control evaluation rule so as to output a quality control evaluation result of the DR image to be evaluated, and specifically includes the following steps:
according to the target part corresponding to the DR image, a plurality of quality control tasks are preset, wherein each quality control task comprises: quality control standards, involved parts and judgment conditions;
and calculating the score value of each quality control standard corresponding to the DR image to be evaluated according to the prediction result of the DR image prediction module on the DR image to be evaluated and the judgment condition of the quality control task, and obtaining the quality control evaluation result of the DR image to be evaluated through weighting calculation.
10. The medical image quality control method based on the computer vision technology as claimed in claim 1, wherein: further comprising:
constructing a basic information integrity evaluation model;
extracting basic information of an inspection item of the medical image data to be evaluated, wherein the basic information at least comprises: basic information, medical history, examination part, examination method, examination item name and unique hospital identification code of the examined person;
predicting the basic information of the examination item of the medical image data to be evaluated based on the basic information integrity evaluation model to obtain a prediction result; and scoring the prediction result according to the preset medical image quality control evaluation rule to obtain a basic information integrity scoring result.
11. The medical image quality control method based on computer vision technology of claim 10, wherein: and automatically generating a medical image quality control evaluation report corresponding to the medical image data to be evaluated by combining the basic information integrity evaluation result and the DR image quality control evaluation result.
12. A medical image quality control system based on computer vision technology, the medical image comprises a DR image, and the medical image quality control system is characterized by comprising:
the acquisition module is used for acquiring a plurality of historical image inspection data of the medical institution within the medical image mutual-recognition area range within a preset time period;
a preprocessing module: the preprocessing module is used for extracting DR images in the historical image inspection data, preprocessing the DR images and generating marked DR image sets with the same angle;
a model generation module: the model generation module utilizes a computer vision technology to construct an initial DR image recognition model for quality control evaluation, and trains, verifies and tests the initial DR image recognition model based on the DR image set to generate a DR image prediction module;
DR image quality control module: the quality control module acquires a DR image to be evaluated based on medical image data to be evaluated, and inputs the DR image to be evaluated into the DR image prediction module to acquire a prediction result; the prediction result comprises various indexes of image quality control; and scoring each index based on a preset medical image quality control evaluation rule, so as to output a quality control evaluation result of the DR image to be evaluated.
13. The medical image quality control system based on computer vision technology of claim 12, wherein: further comprising:
a model construction module: the model construction module is used for constructing a basic information integrity evaluation model;
the information extraction module: the information extraction module is used for extracting basic information of the examination item of the medical image data to be evaluated, and the basic information at least comprises: basic information, medical history, examination part, examination method, examination item name and unique hospital identification code of the examined person;
basic information integrity quality control module: the basic information integrity quality control module is used for predicting the basic information of the examination item of the medical image data to be evaluated based on the basic information integrity evaluation model to obtain a prediction result; and scoring the prediction result according to the preset medical image quality control evaluation rule to obtain a basic information integrity scoring result.
14. The medical image quality control system based on computer vision technology of claim 13, wherein: the medical image quality control report generation module is used for automatically generating a medical image quality control evaluation report corresponding to the medical image data to be evaluated according to the evaluation result generated by the basic information integrity degree quality control module and the evaluation result generated by the DR image quality control module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211318143.8A CN115762721A (en) | 2022-10-26 | 2022-10-26 | Medical image quality control method and system based on computer vision technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211318143.8A CN115762721A (en) | 2022-10-26 | 2022-10-26 | Medical image quality control method and system based on computer vision technology |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115762721A true CN115762721A (en) | 2023-03-07 |
Family
ID=85353324
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211318143.8A Pending CN115762721A (en) | 2022-10-26 | 2022-10-26 | Medical image quality control method and system based on computer vision technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115762721A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117275644A (en) * | 2023-08-31 | 2023-12-22 | 广州零端科技有限公司 | Detection result mutual recognition method, system and storage medium based on deep learning |
-
2022
- 2022-10-26 CN CN202211318143.8A patent/CN115762721A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117275644A (en) * | 2023-08-31 | 2023-12-22 | 广州零端科技有限公司 | Detection result mutual recognition method, system and storage medium based on deep learning |
| CN117275644B (en) * | 2023-08-31 | 2024-04-16 | 广州零端科技有限公司 | Detection result mutual recognition method, system and storage medium based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109544518B (en) | Method and system applied to bone maturity assessment | |
| CN110796199B (en) | Image processing method and device and electronic medical equipment | |
| Salman et al. | Classification of real and fake human faces using deep learning | |
| CN111242948B (en) | Image processing method, image processing device, model training method, model training device, image processing equipment and storage medium | |
| CN110827236B (en) | Brain tissue layering method, device and computer equipment based on neural network | |
| WO2020224433A1 (en) | Target object attribute prediction method based on machine learning and related device | |
| CN111192660B (en) | Image report analysis method, device and computer storage medium | |
| CN111127400A (en) | Method and device for detecting breast lesions | |
| CN116129141A (en) | Medical data processing method, apparatus, device, medium and computer program product | |
| CN114663426A (en) | Bone age assessment method based on key bone area positioning | |
| Adegun et al. | Deep learning model for skin lesion segmentation: Fully convolutional network | |
| CN116883768A (en) | Lung nodule intelligent grading method and system based on multi-modal feature fusion | |
| CN115762721A (en) | Medical image quality control method and system based on computer vision technology | |
| KR20200084816A (en) | Method, apparatus and computer program for analyzing new contents for solving cold start | |
| CN115036034B (en) | Similar patient identification method and system based on patient characterization map | |
| CN116778579A (en) | Multi-person gesture recognition method and device, storage medium and electronic equipment | |
| He et al. | Midcn: A multiple instance deep convolutional network for image classification | |
| CN111582404B (en) | Content classification method, device and readable storage medium | |
| Perkonigg et al. | Detecting bone lesions in multiple myeloma patients using transfer learning | |
| CN113643283A (en) | Method, device, equipment and storage medium for detecting aging condition of human body | |
| CN113822846A (en) | Method, apparatus, device and medium for determining region of interest in medical image | |
| CN115831356B (en) | Auxiliary prediction diagnosis method based on artificial intelligence algorithm | |
| CN117095241B (en) | Screening method, system, equipment and medium for drug-resistant phthisis class | |
| CN115810016B (en) | Automatic identification method, system, storage medium and terminal for CXR (Lung infection) image | |
| CN116563524B (en) | Glance path prediction method based on multi-vision memory unit |
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 |