Automatic analysis and labeling system of medical images based on deep learning

Abstract

The present invention provides an automatic analysis and labeling system for medical images based on deep learning. By adopting a deep learning model, based on historical medical images, multiple medical image recognition models are trained to obtain a set of image recognition models, and applicable image features of each medical image recognition model are determined. Based on the association between the target image features of the target medical image and the applicable image features, multiple medical image recognition models are selected from the image recognition model set to identify and label the target medical image, and a labeled area of ​​the target medical image is obtained. Through multi-model recognition and labeling, and through automatic analysis and labeling of medical images, the workload of doctors can be greatly reduced, the diagnostic efficiency can be improved, and the model accuracy can be improved. Finally, the labeled area is automatically diagnosed to obtain the lesion characteristics and diagnostic suggestions of the target medical image, thereby improving the work efficiency and diagnostic accuracy of doctors and reducing human errors.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an automatic analysis and marking system for medical images based on deep learning. Background Art

[0002] In the field of medical imaging, traditional medical image analysis mainly relies on the expertise and experience of doctors. However, due to the complexity and diversity of medical image information, doctors may be affected by various factors during the analysis and diagnosis process, resulting in inaccurate diagnosis results. In addition, manual annotation of medical images is time-consuming, labor-intensive, and prone to errors.

[0003] In recent years, deep learning technology has made significant progress in the field of medical image processing. By training a large amount of medical image data, deep learning models can learn the characteristics and laws in medical images and realize automatic analysis and diagnosis of medical images. However, the existing deep learning-based automatic analysis and labeling systems for medical images still have some problems, such as the generalization ability of the model is not strong and the accuracy of labeling needs to be improved. Summary of the invention

[0004] The present invention provides a deep learning medical image automatic analysis and labeling system to solve the problems raised in the background technology.

[0005] A deep learning-based automatic analysis and labeling system for medical images, comprising:

[0006] A model design module, for using a deep learning model to train multiple medical image recognition models based on historical medical image images, obtain an image recognition model set, and determine applicable image features of each medical image recognition model;

[0007] An image recognition and annotation module, for selecting a plurality of medical image recognition models from an image recognition model set to recognize and annotate the target medical image based on the association between the target image features and the applicable image features of the target medical image, and obtaining an annotated area of ​​the target medical image;

[0008] The diagnosis module is used to automatically diagnose the marked area and obtain the lesion characteristics and diagnosis suggestions of the target medical image.

[0009] Preferably, it also includes: a user interaction module;

[0010] The user interaction module comprises:

[0011] An interface unit, used to summarize the lesion characteristics and diagnostic suggestions of the target medical image to generate a processing result catalog for interactive interface display;

[0012] The selection unit is used for the user to click and select the corresponding processing result according to the processing result catalog, and the specific content of the processing result is displayed.

[0013] Preferably, the model design module includes:

[0014] A training unit, used to obtain different types of deep learning models, and use historical medical image images to train each deep learning model to obtain multiple medical image recognition models, and construct multiple medical image recognition models to obtain an image recognition model set;

[0015] The matching unit is used to match the corresponding applicable image features for the medical image recognition model based on the training results of the medical image recognition model.

[0016] Preferably, the training unit comprises:

[0017] A division unit, used for dividing the historical medical image images into multiple groups of medical image images according to the lesion type, and marking the medical image images based on the lesion characteristics to obtain multiple groups of initial training images;

[0018] A training unit is used to train different types of deep learning models with multiple groups of initial training images, obtain the training effects of different initial training images on different types of deep learning models, and match the optimal target deep learning model for each group of initial training images based on the training effects;

[0019] A secondary training unit, configured to perform enhancement processing on the initial training image to obtain a second training image, and perform secondary training on the target deep learning model based on the second training image to obtain an initial medical image recognition model;

[0020] An evaluation unit, configured to determine a plurality of access positions based on the model structure of the initial medical image recognition model, respectively add attention mechanism modules to the access positions to obtain a plurality of intermediate medical image recognition models, evaluate the model accuracy of the plurality of intermediate medical image recognition models using historical medical images, and select the intermediate medical image recognition model with the highest accuracy as the final medical image recognition model;

[0021] The integration unit is used to integrate the medical image recognition models to obtain an image recognition model set.

[0022] Preferably, the matching unit comprises:

[0023] An acquisition unit, used to acquire an initial training image and a secondary training image corresponding to the medical image recognition model;

[0024] A feature extraction unit, used to extract features from the initial training image to obtain basic image features, and to extract features from the secondary training image to obtain additional image features;

[0025] The integration unit is used to integrate the basic image features and the additional image features to obtain applicable image features.

[0026] Preferably, it also includes: an image processing module, used to process the collected initial medical image to obtain a target medical image;

[0027] The image processing module comprises:

[0028] A denoising unit, used for filtering the initial medical image to obtain a denoised medical image;

[0029] The normalization unit is used to perform normalization processing on the denoised medical image to obtain a target medical image.

[0030] Preferably, the image recognition and annotation module includes:

[0031] A feature matching unit is used to extract features from a target medical image to obtain target image features, perform feature enhancement on the target image features to obtain enhanced image features, match the target image features with applicable image features to obtain a first correlation degree, and match the enhanced image features with applicable image features to obtain a second correlation degree;

[0032] A selection unit, configured to select a medical image recognition model having a first correlation degree greater than a first preset correlation degree and a second correlation degree greater than a second preset correlation degree from the image recognition model set to recognize the target medical image and obtain a recognition output result;

[0033] The recognition and annotation unit is used to fuse and optimize the recognition output results to obtain a target recognition result, and to annotate the target medical image according to the target recognition result to obtain an annotated area of ​​the target medical image.

[0034] Preferably, the image recognition and annotation unit includes:

[0035] A comparison unit, used for comparing all the recognition output results, obtaining all the same first image regions in all the recognition output results, and using other regions of the target medical image as second image regions;

[0036] A judging unit, configured to obtain image region features of the second image region in each recognition output result, obtain difference features of the image region features, and judge whether the attributes of the difference features are unique;

[0037] If so, use the difference feature as a fusion feature;

[0038] Otherwise, the difference features with the same attributes are taken as a feature group, and the average feature value in a feature group is taken as the fusion feature;

[0039] A fusion unit, used for fusing the image features and fusion features of the first image region to obtain initial region features;

[0040] an optimization unit, configured to determine an optimal reference image feature from the historical medical image based on the relationship between the historical recognition results of the historical medical image and the lesion recognition accuracy, and optimize the initial region feature using the optimal reference image feature to obtain a target recognition region feature;

[0041] The labeling unit is used to label the target recognition area to obtain the labeling area of ​​the target medical image.

[0042] Preferably, the marking unit includes:

[0043] A method determination unit, used for obtaining a lesion area from a target recognition area, and matching a corresponding annotation method based on a shape and a size of the lesion area;

[0044] The region marking unit is used to mark the lesion region according to the marking method to obtain a marked region.

[0045] Preferably, the diagnostic module comprises:

[0046] A lesion determination unit, used for inputting the marked area into an automatic diagnosis model and outputting lesion features of a target medical image;

[0047] The suggestion determination unit is used to match the lesion characteristics with the diagnosis scheme based on the correspondence between the lesion characteristics and the diagnosis scheme, obtain multiple diagnosis schemes, integrate the diagnosis schemes, and obtain diagnosis suggestions.

[0048] Compared with the prior art, the present invention has achieved the following beneficial effects:

[0049] By adopting a deep learning model, based on historical medical images, multiple medical image recognition models are trained to obtain an image recognition model set, and the applicable image features of each medical image recognition model are determined. Based on the association between the target image features and the applicable image features of the target medical image, multiple medical image recognition models are selected from the image recognition model set to identify and annotate the target medical image, and the annotated area of ​​the target medical image is obtained. Through multi-model recognition and annotation, by automatically analyzing and annotating medical images, the workload of doctors can be greatly reduced, the diagnostic efficiency can be improved, and the model accuracy can be improved. Finally, the annotated area is automatically diagnosed to obtain the lesion characteristics and diagnostic suggestions of the target medical image, improve the work efficiency and diagnostic accuracy of doctors, and reduce human errors. The system has broad application prospects and market demand, and has important practical and economic value.

[0050] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0051] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0053] Figure 1 This is a structural diagram of a medical image automatic analysis and marking system based on deep learning in an embodiment of the present invention;

[0054] Figure 2 A structural diagram of the model design module described in an embodiment of the present invention;

[0055] Figure 3 4 is a structural diagram of the image recognition and annotation module described in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0057] Embodiment 1:

[0058] The present invention provides a deep learning-based medical image automatic analysis and labeling system. Figure 1 As shown, including:

[0059] A model design module, for using a deep learning model to train multiple medical image recognition models based on historical medical image images, obtain an image recognition model set, and determine applicable image features of each medical image recognition model;

[0060] An image recognition and annotation module, for selecting a plurality of medical image recognition models from an image recognition model set to recognize and annotate the target medical image based on the association between the target image features and the applicable image features of the target medical image, and obtaining an annotated area of ​​the target medical image;

[0061] The diagnosis module is used to automatically diagnose the marked area and obtain the lesion characteristics and diagnosis suggestions of the target medical image.

[0062] In this embodiment, the deep learning model is, for example, a deep convolutional model, a machine learning model in a neural network model.

[0063] In this embodiment, a plurality of medical imaging image recognition models are selected from the image recognition model set to select target image features with a high correlation with applicable image features.

[0064] The beneficial effects of the above design scheme are: by adopting a deep learning model, based on historical medical image images, multiple medical image recognition models are trained to obtain a set of image recognition models, and the applicable image features of each medical image recognition model are determined. Based on the association between the target image features of the target medical image and the applicable image features, multiple medical image recognition models are selected from the set of image recognition models to identify and annotate the target medical image, and the annotated area of ​​the target medical image is obtained. Through multi-model recognition and annotation, by automatically analyzing and annotating medical images, the workload of doctors can be greatly reduced, the diagnostic efficiency can be improved, and the model accuracy can be improved. Finally, the annotated area is automatically diagnosed to obtain the lesion characteristics and diagnostic suggestions of the target medical image, improve the work efficiency and diagnostic accuracy of doctors, and reduce human errors. The system has broad application prospects and market demand, and has important practical and economic value.

[0065] Embodiment 2:

[0066] Based on Example 1, the embodiment of the present invention provides a medical image automatic analysis and marking system based on deep learning, further comprising: a user interaction module;

[0067] The user interaction module comprises:

[0068] An interface unit, used to summarize the lesion characteristics and diagnostic suggestions of the target medical image to generate a processing result catalog for interactive interface display;

[0069] The selection unit is used for the user to click and select the corresponding processing result according to the processing result catalog, and the specific content of the processing result is displayed.

[0070] The beneficial effect of the above design is: providing a user interaction interface to facilitate users to view and analyze the processing results of medical images. Users can select different medical images for analysis and annotation through the interface, and can view detailed diagnosis reports.

[0071] Embodiment 3:

[0072] Based on Example 1, the present invention provides a medical image automatic analysis and marking system based on deep learning, such as Figure 2 As shown, the model design module includes:

[0073] A training unit, used to obtain different types of deep learning models, and use historical medical image images to train each deep learning model to obtain multiple medical image recognition models, and construct multiple medical image recognition models to obtain an image recognition model set;

[0074] The matching unit is used to match the corresponding applicable image features for the medical image recognition model based on the training results of the medical image recognition model.

[0075] The beneficial effect of the above design scheme is: based on the training results of the medical image recognition model, the corresponding applicable image features are matched to the medical image recognition model, providing a feature basis for selecting a suitable model for medical images.

[0076] Embodiment 4:

[0077] Based on Example 3, the embodiment of the present invention provides a medical image automatic analysis and labeling system based on deep learning, wherein the training unit includes:

[0078] A division unit, used for dividing the historical medical image images into multiple groups of medical image images according to the lesion type, and marking the medical image images based on the lesion characteristics to obtain multiple groups of initial training images;

[0079] A training unit is used to train different types of deep learning models with multiple groups of initial training images, obtain the training effects of different initial training images on different types of deep learning models, and match the optimal target deep learning model for each group of initial training images based on the training effects;

[0080] A secondary training unit, configured to perform enhancement processing on the initial training image to obtain a second training image, and perform secondary training on the target deep learning model based on the second training image to obtain an initial medical image recognition model;

[0081] An evaluation unit, configured to determine a plurality of access positions based on the model structure of the initial medical image recognition model, respectively add attention mechanism modules to the access positions to obtain a plurality of intermediate medical image recognition models, evaluate the model accuracy of the plurality of intermediate medical image recognition models using historical medical images, and select the intermediate medical image recognition model with the highest accuracy as the final medical image recognition model;

[0082] The integration unit is used to integrate the medical image recognition models to obtain an image recognition model set.

[0083] In this embodiment, the initial training image is enhanced, including operations such as rotation, translation, and scaling, to increase the robustness and generalization ability of the model.

[0084] The beneficial effects of the above design scheme are: increasing the robustness and generalization ability of the model through image enhancement, and by adding an attention mechanism module, the model can focus on important areas in medical images, improving the recognition accuracy of the model, combining the advantages of multiple deep learning models, and using a multi-model fusion method to improve the performance of the model. By fusing and optimizing the output results of different models, more accurate annotation and diagnosis results are obtained, and finally a collection of image recognition models is obtained, providing a model basis for multi-model recognition of image features.

[0085] Embodiment 5:

[0086] Based on Example 4, the embodiment of the present invention provides a medical image automatic analysis and marking system based on deep learning, wherein the matching unit includes:

[0087] An acquisition unit, used to acquire an initial training image and a secondary training image corresponding to the medical image recognition model;

[0088] A feature extraction unit, used to extract features from the initial training image to obtain basic image features, and to extract features from the secondary training image to obtain additional image features;

[0089] The integration unit is used to integrate the basic image features and the additional image features to obtain applicable image features.

[0090] The beneficial effects of the above design scheme are: by performing feature extraction on the initial training image to obtain basic image features, performing feature extraction on the secondary training image to obtain additional image features, integrating the basic image features and the additional image features to obtain applicable image features, ensuring the accuracy and comprehensiveness of the applicable image features obtained, and providing a basis for model matching selection.

[0091] Embodiment 6:

[0092] Based on Example 1, the embodiment of the present invention provides a medical image automatic analysis and marking system based on deep learning, further comprising: an image processing module, for processing the collected initial medical image to obtain a target medical image;

[0093] The image processing module comprises:

[0094] A denoising unit, used for filtering the initial medical image to obtain a denoised medical image;

[0095] The normalization unit is used to perform normalization processing on the denoised medical image to obtain a target medical image.

[0096] The beneficial effect of the above design scheme is: by processing the collected initial medical image, the target medical image is obtained, the image quality is improved, and a high-quality data basis is provided for lesion identification.

[0097] Embodiment 7:

[0098] Based on Example 1, the present invention provides a medical image automatic analysis and marking system based on deep learning, such as Figure 3 As shown, the image recognition and annotation module includes:

[0099] A feature matching unit is used to extract features from a target medical image to obtain target image features, perform feature enhancement on the target image features to obtain enhanced image features, match the target image features with applicable image features to obtain a first correlation degree, and match the enhanced image features with applicable image features to obtain a second correlation degree;

[0100] A selection unit, configured to select a medical image recognition model having a first correlation degree greater than a first preset correlation degree and a second correlation degree greater than a second preset correlation degree from the image recognition model set to recognize the target medical image and obtain a recognition output result;

[0101] The recognition and annotation unit is used to fuse and optimize the recognition output results to obtain a target recognition result, and to annotate the target medical image according to the target recognition result to obtain an annotated area of ​​the target medical image.

[0102] The beneficial effects of the above design scheme are: by performing feature enhancement on the target image features to obtain enhanced image features, matching the target image features with applicable image features to obtain a first degree of association, matching the enhanced image features with applicable image features to obtain a second degree of association, providing comprehensive feature matching for model selection, ensuring the adaptability of the selected model, fusing and optimizing the recognition output results to obtain target recognition results, and annotating the target medical image according to the target recognition results to obtain the annotated area of ​​the target medical image, and automatically analyzing and annotating medical images, the workload of doctors can be greatly reduced and the diagnostic efficiency can be improved.

[0103] Embodiment 8:

[0104] Based on Example 7, the embodiment of the present invention provides a medical image automatic analysis and marking system based on deep learning, wherein the image recognition and marking unit includes:

[0105] A comparison unit, used for comparing all the recognition output results, obtaining all the same first image regions in all the recognition output results, and using other regions of the target medical image as second image regions;

[0106] A judging unit, configured to obtain image region features of the second image region in each recognition output result, obtain difference features of the image region features, and judge whether the attributes of the difference features are unique;

[0107] If so, use the difference feature as a fusion feature;

[0108] Otherwise, the difference features with the same attributes are taken as a feature group, and the average feature value in a feature group is taken as the fusion feature;

[0109] A fusion unit, used for fusing the image features and fusion features of the first image region to obtain initial region features;

[0110] an optimization unit, configured to determine an optimal reference image feature from the historical medical image based on the relationship between the historical recognition results of the historical medical image and the lesion recognition accuracy, and optimize the initial region feature using the optimal reference image feature to obtain a target recognition region feature;

[0111] The labeling unit is used to label the target recognition area to obtain the labeling area of ​​the target medical image.

[0112] The beneficial effect of the above design scheme is as follows: by comparing all the recognition output results, all the same first image regions in all the recognition output results are obtained, other regions of the target medical image are used as second image regions, image region features of the second image region under each recognition output result are obtained, and difference features of the image region features are obtained, and it is determined whether the attributes of the difference features are individually unique. If they are unique, they represent the unique features of the model recognition, otherwise, they represent that the same attribute features recognized by different models are different, and the difference features with the same attributes are used as a feature group, and the average feature value in a feature group is taken as a fusion feature, and the image features and fusion features of the first image region are fused to obtain the initial region features, so as to achieve the fusion of multi-model results and ensure the comprehensiveness and accuracy of the features of the obtained initial region features, and then based on the relationship between the historical recognition results of the historical medical image images and the lesion recognition accuracy, the optimal reference image features are determined from the historical medical image images, and the initial region features are optimized by using the optimal reference image features to obtain the target recognition region features, so as to further improve the accuracy of the obtained target recognition region features, and the target recognition region is annotated to obtain the annotated region of the target medical image, so as to ensure the annotation effect.

[0113] Embodiment 9:

[0114] Based on Example 8, the embodiment of the present invention provides a medical image automatic analysis and marking system based on deep learning, wherein the marking unit includes:

[0115] A method determination unit, used for obtaining a lesion area from a target recognition area, and matching a corresponding annotation method based on a shape and a size of the lesion area;

[0116] The region marking unit is used to mark the lesion region according to the marking method to obtain a marked region.

[0117] The beneficial effect of the above design scheme is that by automatically analyzing and annotating medical images, the workload of doctors can be greatly reduced and the diagnosis efficiency can be improved.

[0118] Embodiment 10:

[0119] Based on Example 1, the embodiment of the present invention provides a medical image automatic analysis and marking system based on deep learning, wherein the diagnosis module includes:

[0120] A lesion determination unit, used for inputting the marked area into an automatic diagnosis model and outputting lesion features of a target medical image;

[0121] The suggestion determination unit is used to match the lesion characteristics with the diagnosis scheme based on the correspondence between the lesion characteristics and the diagnosis scheme, obtain multiple diagnosis schemes, integrate the diagnosis schemes, and obtain diagnosis suggestions.

[0122] The beneficial effect of the above design scheme is that the medical image can be automatically diagnosed based on the annotation results and the output of the deep learning model. The diagnosis module can identify the lesion area in the medical image and give corresponding diagnosis suggestions.

[0123] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention belong to the scope of this application document and its equivalent technology, the present invention is also intended to include these changes and variations.

Claims

1. A medical image automatic analysis and labeling system based on deep learning, characterized in that: include: A model design module, for using a deep learning model to train multiple medical image recognition models based on historical medical image images, obtain an image recognition model set, and determine applicable image features of each medical image recognition model; An image recognition and annotation module, for selecting a plurality of medical image recognition models from an image recognition model set to recognize and annotate the target medical image based on the association between the target image features and the applicable image features of the target medical image, and obtaining an annotated area of ​​the target medical image; The diagnosis module is used to automatically diagnose the marked area and obtain the pathological features and diagnosis suggestions of the target medical image; The model design module includes: A training unit, used to obtain different types of deep learning models, and use historical medical image images to train each deep learning model to obtain multiple medical image recognition models, and construct multiple medical image recognition models to obtain an image recognition model set; A matching unit, for matching corresponding applicable image features for the medical image recognition model based on the training results of the medical image recognition model; The training unit comprises: A division unit, used to divide the historical medical image images into multiple groups of medical image images according to the lesion type, and mark the medical image images based on the lesion characteristics to obtain multiple groups of initial training images; A primary training unit is used to train different types of deep learning models with multiple groups of initial training images, obtain the training effects of different initial training images on different types of deep learning models, and match the optimal target deep learning model for each group of initial training images based on the training effects; A secondary training unit, configured to perform enhancement processing on the initial training image to obtain a second training image, and perform secondary training on the target deep learning model based on the second training image to obtain an initial medical image recognition model; An evaluation unit, configured to determine a plurality of access positions based on the model structure of the initial medical image recognition model, respectively add attention mechanism modules to the access positions to obtain a plurality of intermediate medical image recognition models, evaluate the model accuracy of the plurality of intermediate medical image recognition models using historical medical images, and select the intermediate medical image recognition model with the highest accuracy as the final medical image recognition model; A first integration unit is used to integrate the medical image recognition models to obtain an image recognition model set; The matching unit comprises: An acquisition unit, used to acquire an initial training image and a secondary training image corresponding to the medical image recognition model; A feature extraction unit, used to extract features from the initial training image to obtain basic image features, and to extract features from the secondary training image to obtain additional image features; The second integration unit is used to integrate the basic image features and the additional image features to obtain applicable image features.

2. According to claim 1, a deep learning-based automatic analysis and marking system for medical images, characterized in that: Also includes: User interaction module; The user interaction module comprises: An interface unit, used to summarize the lesion characteristics and diagnostic suggestions of the target medical image to generate a processing result catalog for interactive interface display; The selection unit is used for the user to click and select the corresponding processing result according to the processing result catalog, and the specific content of the processing result is displayed.

3. The medical image automatic analysis and marking system based on deep learning according to claim 1, characterized in that: Also includes: An image processing module is used to process the acquired initial medical image to obtain a target medical image; The image processing module comprises: A denoising unit, used for filtering the initial medical image to obtain a denoised medical image; The normalization unit is used to perform normalization processing on the denoised medical image to obtain a target medical image.

4. The deep learning-based automatic analysis and labeling system for medical images according to claim 1, characterized in that: The image recognition and annotation module comprises: A feature matching unit is used to extract features from a target medical image to obtain target image features, perform feature enhancement on the target image features to obtain enhanced image features, match the target image features with applicable image features to obtain a first correlation degree, and match the enhanced image features with applicable image features to obtain a second correlation degree; A selection unit, configured to select a medical image recognition model having a first correlation degree greater than a first preset correlation degree and a second correlation degree greater than a second preset correlation degree from the image recognition model set to recognize the target medical image and obtain a recognition output result; The recognition and annotation unit is used to fuse and optimize the recognition output results to obtain a target recognition result, and to annotate the target medical image according to the target recognition result to obtain an annotated area of ​​the target medical image.

5. The medical image automatic analysis and marking system based on deep learning according to claim 4, characterized in that: The identification and marking unit comprises: A comparison unit, used for comparing all the recognition output results, obtaining all the same first image regions in all the recognition output results, and using other regions of the target medical image as second image regions; A judging unit, configured to obtain image region features of the second image region in each recognition output result, obtain difference features of the image region features, and judge whether the attributes of the difference features are unique; If so, use the difference feature as a fusion feature; Otherwise, the difference features with the same attributes are taken as a feature group, and the average feature value in a feature group is taken as the fusion feature; A fusion unit, used for fusing the image features and fusion features of the first image region to obtain initial region features; an optimization unit, configured to determine an optimal reference image feature from the historical medical image based on the relationship between the historical recognition results of the historical medical image and the lesion recognition accuracy, and optimize the initial region feature using the optimal reference image feature to obtain a target recognition region feature; The labeling unit is used to label the target recognition area to obtain the labeling area of ​​the target medical image.

6. The deep learning-based automatic analysis and labeling system for medical images according to claim 5, characterized in that: The marking unit comprises: A method determination unit, used for obtaining a lesion area from a target recognition area, and matching a corresponding annotation method based on a shape and a size of the lesion area; The region marking unit is used to mark the lesion region according to the marking method to obtain a marked region.

7. The medical image automatic analysis and marking system based on deep learning according to claim 1, characterized in that: The diagnostic module comprises: A lesion determination unit, used for inputting the marked area into an automatic diagnosis model and outputting lesion features of a target medical image; The suggestion determination unit is used to match the lesion characteristics with the diagnosis scheme based on the correspondence between the lesion characteristics and the diagnosis scheme, obtain multiple diagnosis schemes, integrate the diagnosis schemes, and obtain diagnosis suggestions.

Images