Deep learning-based medical image diagnosis method and system, and medium
Through deep learning-based medical imaging diagnosis methods, including preprocessing, model matching and optimization, the problem of high misdiagnosis rate in the prior art is solved, and higher diagnostic accuracy and adaptability are achieved.
Patent Information
- Application Number
- CN202411810992.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-10
- Publication Date
- 2025-04-29
AI Technical Summary
Misdiagnosis still exists in existing medical imaging diagnostic technologies, affecting patients' health and treatment effects.
The medical image diagnosis method based on deep learning is adopted, and the medical image is preprocessed, the initial image diagnosis model is matched, the reference image is matched based on similarity, and the initial model is trained and optimized by deep learning to obtain the target image diagnosis model, and then medical diagnosis is carried out.
It improves the accuracy of medical imaging diagnosis, ensures the adaptability of the target imaging diagnostic model to the diagnostic images, and reduces the misdiagnosis rate.
Smart Images
Figure CN119905238A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of image processing technology, and in particular to a medical image diagnosis method, system and medium based on deep learning. Background Art
[0002] Medical imaging diagnosis is of great significance in the medical field, which is mainly reflected in the following aspects: providing comprehensive lesion information, assisting in disease diagnosis and treatment, evaluating treatment effects and prognosis, and early detection and treatment. Medical imaging diagnosis plays an irreplaceable role in the medical field. It not only provides an important basis for the diagnosis and treatment of diseases, but also plays a key role in the early detection of diseases, efficacy evaluation and prognosis evaluation.
[0003] Although the development of modern imaging technology has made imaging diagnosis more accurate and reliable, misdiagnosis is still an existing and serious problem, which has a huge impact on patients' health and treatment. Summary of the invention
[0004] The present invention provides a medical imaging diagnosis method, system and medium based on deep learning to solve the problems raised in the background technology.
[0005] A medical imaging diagnosis method based on deep learning includes:
[0006] S1: preprocessing the medical image to obtain the medical image to be diagnosed, and matching the initial image diagnosis model that meets the diagnosis requirements from the image diagnosis model database based on the image type of the medical image to be diagnosed;
[0007] S2: Based on the image features of the medical image to be diagnosed, matching a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity from the image database;
[0008] S3: Based on the reference medical images, the initial image diagnosis model is trained and optimized by using deep learning to obtain a target image diagnosis model;
[0009] S4: Inputting the medical image to be diagnosed into the target image diagnosis model to obtain a medical diagnosis result.
[0010] Preferably, in S1, preprocessing the medical image to obtain the medical image to be diagnosed includes:
[0011] Based on the image type of the medical image, determining the main judgment features of the medical image, and determining specific preprocessing parameters of image denoising, image enhancement, image smoothing and image transformation based on the main judgment features;
[0012] The medical image is preprocessed according to the preprocessing parameters to obtain the medical image to be diagnosed.
[0013] Preferably, in S1, based on the image type of the medical image to be diagnosed, matching an initial image diagnosis model that meets the diagnosis requirements from an image diagnosis model database includes:
[0014] Based on the image type of the medical image to be diagnosed, matching and acquiring a first image diagnosis model group that satisfies the image type from an image diagnosis model database;
[0015] Determine the diagnostic output result characteristics based on the diagnostic requirements of the medical image to be diagnosed, and select a second image diagnostic model group that meets the diagnostic output result characteristics from the first image diagnostic model group;
[0016] Performing preliminary feature extraction on the image to be diagnosed based on a preset basic extraction method to obtain basic features of the image, obtaining training image features of each model in the second image diagnosis model group, matching the target basic features with the training image features, and obtaining a third image diagnosis model group with a matching degree greater than a preset matching degree from the second image diagnosis model group;
[0017] Obtaining the latest update time of each model in the third image diagnosis model group, and selecting a fourth image diagnosis model group whose latest update time meets a preset date;
[0018] Obtaining historical diagnostic data of each model in the fourth image diagnosis model group, obtaining diagnostic image features and diagnostic result features from the historical diagnostic data, determining a first weight based on the similarity between the medical image to be diagnosed and the diagnostic image features, and determining a second weight based on the diagnostic correctness of the diagnostic result features;
[0019] Each model of the fourth image diagnosis model group is evaluated based on the first weight and the second weight to determine an evaluation value, and the model with the largest evaluation value is selected as the initial image diagnosis model.
[0020] Preferably, in S2, based on the image features of the medical image to be diagnosed, matching a reference medical image having a similarity with the medical image to be diagnosed greater than a preset similarity from an image database includes:
[0021] matching an initial medical image of the same image type as the medical image to be diagnosed from an image database;
[0022] Obtaining image black shadow features, image white shadow features and image density features from image features of the medical image to be diagnosed;
[0023] Based on the image black shadow feature, the image white shadow feature and the image density feature, a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity is matched from the initial medical image.
[0024] Preferably, based on the image black shadow feature, the image white shadow feature and the image density feature, matching a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity from the initial medical image includes:
[0025] Based on the image type of the medical image to be diagnosed, determining feature weights for image black shadow features, image white shadow features and image density features respectively;
[0026] Establishing a weighted matching rule based on the feature weights of the image black shadow feature, the image white shadow feature and the image density feature, and determining the similarity between the initial medical image and the medical image to be diagnosed according to the weighted matching rule;
[0027] Matching the initial medical image with a reference medical image whose similarity is greater than a preset similarity.
[0028] Preferably, in S3, based on the reference medical images, the initial image diagnosis model is trained and optimized by using deep learning to obtain the target image diagnosis model, including:
[0029] Acquire multiple identification features of the medical image to be diagnosed, mark the reference medical image based on the multiple identification features to obtain multiple marking features, and determine the marking weight of each marking feature based on the importance of each identification feature of the medical image to be diagnosed to the medical diagnosis;
[0030] Obtain a first training sample based on the reference medical image, and train the initial image diagnosis model using deep learning based on the first training sample to obtain a first intermediate diagnosis model;
[0031] Obtaining a second training sample based on the marking features and marking weights of the reference medical image, and training the initial image diagnosis model using deep learning based on the second training sample to obtain a second intermediate diagnosis model;
[0032] Based on the model structure and model accuracy of the first intermediate diagnostic model and the model structure and model accuracy of the second intermediate diagnostic model, a functional relationship between the model structure and the model accuracy is established;
[0033] Based on the functional relationship, determine the target model structure corresponding to the maximum model accuracy;
[0034] The initial image diagnosis model is optimized based on the target model structure to obtain a target image diagnosis model.
[0035] Preferably, the initial image diagnosis model is optimized based on the target model structure to obtain the target image diagnosis model, including:
[0036] obtaining an initial model structure of an initial imaging diagnosis model, and determining a structural difference between the initial model structure and a target model structure;
[0037] Optimization parameters are determined based on the structural differences, and the initial image diagnosis model is optimized based on the optimization parameters to obtain a target image diagnosis model.
[0038] Preferably, in S4, the medical image to be diagnosed is input into a target image diagnosis model to obtain a medical diagnosis result, including:
[0039] Inputting the medical image to be diagnosed into the target image diagnosis model to obtain an image feature output result of the medical image to be diagnosed;
[0040] The image feature data result is matched with a preset diagnosis correspondence table to obtain a medical diagnosis result.
[0041] A diagnostic system for a medical imaging diagnostic method based on deep learning, comprising:
[0042] The model selection module is used to pre-process the medical images to obtain the medical images to be diagnosed, and based on the image type of the medical images to be diagnosed, to match the initial image diagnosis model that meets the diagnosis requirements from the image diagnosis model database;
[0043] An image selection module, used for matching a reference medical image having a similarity greater than a preset similarity with the medical image to be diagnosed from an image database based on the image features of the medical image to be diagnosed;
[0044] A model optimization module, used to train and optimize the initial image diagnosis model using deep learning based on reference medical images to obtain a target image diagnosis model;
[0045] The medical diagnosis module is used to input the medical image to be diagnosed into the target image diagnosis model to obtain a medical diagnosis result.
[0046] A storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any one of the medical imaging diagnosis methods based on deep learning.
[0047] Compared with the prior art, the present invention has achieved the following beneficial effects:
[0048] By preprocessing the medical images, the medical images to be diagnosed are obtained. Based on the image type of the medical images to be diagnosed, the initial image diagnosis model that meets the diagnosis requirements is matched from the image diagnosis model database to achieve the selection of the diagnosis model type and provide a model basis for medical image diagnosis. Based on the image features of the medical images to be diagnosed, reference medical images whose similarity with the medical images to be diagnosed is greater than a preset similarity are matched from the image database. Based on the reference medical images, the initial image diagnosis model is trained and optimized using deep learning to obtain a target image diagnosis model, and the training optimization of the medical image model based on deep learning is achieved to ensure the adaptability of the target image diagnosis model to the medical images to be diagnosed and the accuracy of the medical diagnosis results obtained by inputting the medical images to the target image diagnosis model into the diagnosis model.
[0049] 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.
[0050] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] 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:
[0052] Figure 1 is a flowchart of a medical imaging diagnosis method based on deep learning in an embodiment of the present invention;
[0053] Figure 2 A flowchart of selecting a reference medical image in an embodiment of the present invention;
[0054] Figure 3 This is a structural diagram of a medical imaging diagnosis system based on deep learning in an embodiment of the present invention. DETAILED DESCRIPTION
[0055] 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.
[0056] Embodiment 1:
[0057] The present invention provides a medical imaging diagnosis method based on deep learning. Figure 1 As shown, including:
[0058] S1: preprocessing the medical image to obtain the medical image to be diagnosed, and matching the initial image diagnosis model that meets the diagnosis requirements from the image diagnosis model database based on the image type of the medical image to be diagnosed;
[0059] S2: Based on the image features of the medical image to be diagnosed, matching a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity from the image database;
[0060] S3: Based on the reference medical images, the initial image diagnosis model is trained and optimized by using deep learning to obtain a target image diagnosis model;
[0061] S4: Inputting the medical image to be diagnosed into the target image diagnosis model to obtain a medical diagnosis result.
[0062] In this embodiment, the preprocessing of the medical image includes image denoising, image enhancement, image smoothing and image transformation.
[0063] In this embodiment, the types of medical images to be diagnosed include vascular subtraction, magnetic resonance imaging, and ultrasound imaging.
[0064] In this embodiment, the image features of the medical image to be diagnosed include structural features, shadow features, density features, etc.
[0065] In this embodiment, the image diagnosis model collects various comprehensive diagnosis model databases formed by various types of medical images and diagnosis requirements.
[0066] In this embodiment, the image database collects various types of medical images.
[0067] The beneficial effects of the above design scheme are: by preprocessing the medical image, the medical image to be diagnosed is obtained; based on the image type of the medical image to be diagnosed, the initial image diagnosis model that meets the diagnosis requirements is matched from the image diagnosis model database to achieve the selection of the diagnosis model type, and provide a model basis for medical image diagnosis; based on the image features of the medical image to be diagnosed, a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity is matched from the image database; based on the reference medical image, the initial image diagnosis model is trained and optimized using deep learning to obtain a target image diagnosis model, and the training optimization of the medical image model based on deep learning is achieved, so as to ensure the adaptability of the target image diagnosis model to the medical image to be diagnosed, and ensure the accuracy of the medical diagnosis result obtained by inputting the medical image to be diagnosed into the target image diagnosis model.
[0068] Embodiment 2:
[0069] Based on Example 1, the embodiment of the present invention provides a medical image diagnosis method based on deep learning. In S1, the medical image is preprocessed to obtain a medical image to be diagnosed, including:
[0070] Based on the image type of the medical image, determining the main judgment features of the medical image, and determining specific preprocessing parameters of image denoising, image enhancement, image smoothing and image transformation based on the main judgment features;
[0071] The medical image is preprocessed according to the preprocessing parameters to obtain the medical image to be diagnosed.
[0072] In this embodiment, the main judgment features are, for example, shadow area, white spot features, etc.
[0073] The beneficial effect of the above design scheme is: by determining the main judgment features of the medical image based on the image type of the medical image, and determining the specific preprocessing parameters of image denoising, image enhancement, image smoothing and image transformation based on the main judgment features, the medical image is preprocessed according to the preprocessing parameters to obtain the medical image to be diagnosed, and high-quality images are provided for medical image diagnosis.
[0074] Embodiment 3:
[0075] Based on Example 1, the embodiment of the present invention provides a medical image diagnosis method based on deep learning. In S1, based on the image type of the medical image to be diagnosed, an initial image diagnosis model that meets the diagnosis requirements is matched from an image diagnosis model database, including:
[0076] Based on the image type of the medical image to be diagnosed, matching and acquiring a first image diagnosis model group that satisfies the image type from an image diagnosis model database;
[0077] Determine the diagnostic output result characteristics based on the diagnostic requirements of the medical image to be diagnosed, and select a second image diagnostic model group that meets the diagnostic output result characteristics from the first image diagnostic model group;
[0078] Performing preliminary feature extraction on the image to be diagnosed based on a preset basic extraction method to obtain basic features of the image, obtaining training image features of each model in the second image diagnosis model group, matching the target basic features with the training image features, and obtaining a third image diagnosis model group with a matching degree greater than a preset matching degree from the second image diagnosis model group;
[0079] Obtaining the latest update time of each model in the third image diagnosis model group, and selecting a fourth image diagnosis model group whose latest update time meets a preset date;
[0080] Obtaining historical diagnostic data of each model in the fourth image diagnosis model group, obtaining diagnostic image features and diagnostic result features from the historical diagnostic data, determining a first weight based on the similarity between the medical image to be diagnosed and the diagnostic image features, and determining a second weight based on the diagnostic correctness of the diagnostic result features;
[0081] Each model of the fourth image diagnosis model group is evaluated based on the first weight and the second weight to determine an evaluation value, and the model with the largest evaluation value is selected as the initial image diagnosis model.
[0082] In this embodiment, the first weight and the second weight are used to evaluate each model of the fourth imaging diagnosis model group and are specifically designed according to actual conditions. For example, the evaluation value is the sum of the first weight and the second weight.
[0083] The beneficial effect of the above design scheme is: by mechanically screening the medical diagnosis model from five aspects: image novelty, diagnostic needs, image characteristics, timeliness and historical diagnosis, the initial image diagnosis model is finally obtained, the accurate selection of the diagnosis model type is achieved, and a model basis is provided for medical image diagnosis.
[0084] Embodiment 4:
[0085] Based on Example 1, the present invention provides a medical imaging diagnosis method based on deep learning, such as Figure 2 As shown, in S2, based on the image features of the medical image to be diagnosed, matching a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity from the image database includes:
[0086] matching an initial medical image of the same image type as the medical image to be diagnosed from an image database;
[0087] Obtaining image black shadow features, image white shadow features and image density features from image features of the medical image to be diagnosed;
[0088] Based on the image black shadow feature, the image white shadow feature and the image density feature, a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity is matched from the initial medical image.
[0089] The beneficial effect of the above design scheme is: by matching the initial medical image of the same image type as the medical image to be diagnosed from the image database, the image black shadow feature, the image white shadow feature and the image density feature are obtained from the image features of the medical image to be diagnosed, and based on the image black shadow feature, the image white shadow feature and the image density feature, a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity is matched from the initial medical image, thereby providing a training data basis for the training optimization of the medical imaging model.
[0090] Embodiment 5:
[0091] Based on Example 4, the embodiment of the present invention provides a medical image diagnosis method based on deep learning, which matches a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity from an initial medical image based on image black shadow features, image white shadow features and image density features, including:
[0092] Based on the image type of the medical image to be diagnosed, determining feature weights for image black shadow features, image white shadow features and image density features respectively;
[0093] Establishing a weighted matching rule based on the feature weights of the image black shadow feature, the image white shadow feature and the image density feature, and determining the similarity between the initial medical image and the medical image to be diagnosed according to the weighted matching rule;
[0094] Matching the initial medical image with a reference medical image whose similarity is greater than a preset similarity.
[0095] The beneficial effect of the above design scheme is: by determining the feature weights of the image black shadow feature, the image white shadow feature and the image density feature respectively based on the image type of the medical image to be diagnosed, and establishing a weighted matching rule based on the feature weights of the image black shadow feature, the image white shadow feature and the image density feature, according to the weighted matching rule, determining the similarity between the initial medical image and the medical image to be diagnosed, matching the reference medical image whose similarity is greater than the preset similarity from the initial medical image, and providing a training data basis for the training optimization of the medical imaging model.
[0096] Embodiment 6:
[0097] Based on Example 1, the embodiment of the present invention provides a medical image diagnosis method based on deep learning. In S3, based on the reference medical image, the initial image diagnosis model is trained and optimized by deep learning to obtain a target image diagnosis model, including:
[0098] Acquire multiple identification features of the medical image to be diagnosed, mark the reference medical image based on the multiple identification features to obtain multiple marking features, and determine the marking weight of each marking feature based on the importance of each identification feature of the medical image to be diagnosed to the medical diagnosis;
[0099] Obtain a first training sample based on the reference medical image, and train the initial image diagnosis model using deep learning based on the first training sample to obtain a first intermediate diagnosis model;
[0100] Obtaining a second training sample based on the marking features and marking weights of the reference medical image, and training the initial image diagnosis model using deep learning based on the second training sample to obtain a second intermediate diagnosis model;
[0101] Based on the model structure and model accuracy of the first intermediate diagnostic model and the model structure and model accuracy of the second intermediate diagnostic model, a functional relationship between the model structure and the model accuracy is established;
[0102] Based on the functional relationship, determine the target model structure corresponding to the maximum model accuracy;
[0103] The initial image diagnosis model is optimized based on the target model structure to obtain a target image diagnosis model.
[0104] In this embodiment, the identification feature corresponds to the marking feature one by one, and the identification feature is, for example, a grayscale feature, an area feature, a shadow feature, etc.
[0105] In this embodiment, the greater the importance of each identification feature to the medical diagnosis, the greater the corresponding label weight.
[0106] In this embodiment, the first intermediate diagnostic model is obtained by training with unlabeled features, and the second intermediate arbitrary model is obtained by training with labeled features.
[0107] The beneficial effect of the above design scheme is: by obtaining multiple identification features of the medical image to be diagnosed, marking the reference medical image based on the multiple identification features to obtain multiple marking features, and determining the marking weight of each marking feature based on the importance of each identification feature of the medical image to be diagnosed to the medical diagnosis, obtaining a first training sample based on the reference medical image, training the initial image diagnosis model using deep learning based on the first training sample to obtain a first intermediate diagnosis model, obtaining a second training sample based on the marking features and marking weights of the reference medical image, training the initial image diagnosis model using deep learning based on the second training sample to obtain a second intermediate diagnosis model, establishing a functional relationship between the model structure and the model accuracy based on the model structure and model accuracy of the first intermediate diagnosis model and the model structure and model accuracy of the second intermediate diagnosis model, determining a target model structure corresponding to the maximum model accuracy based on the functional relationship, optimizing the initial image diagnosis model based on the target model structure to obtain a target image diagnosis model, realizing medical image model training optimization based on deep learning, ensuring the adaptability of the obtained target image diagnosis model to the medical image to be diagnosed, and ensuring the accuracy of the medical diagnosis result obtained by inputting the medical image to be diagnosed into the target image diagnosis model.
[0108] Embodiment 7:
[0109] Based on Example 6, the embodiment of the present invention provides a medical image diagnosis method based on deep learning, which optimizes the initial image diagnosis model based on the target model structure to obtain a target image diagnosis model, including:
[0110] obtaining an initial model structure of an initial imaging diagnosis model, and determining a structural difference between the initial model structure and a target model structure;
[0111] Optimization parameters are determined based on the structural differences, and the initial image diagnosis model is optimized based on the optimization parameters to obtain a target image diagnosis model.
[0112] The beneficial effect of the above design scheme is: by obtaining the initial model structure of the initial image diagnosis model, determining the structural difference between the initial model structure and the target model structure, determining the optimization parameters based on the structural difference, optimizing the initial image diagnosis model based on the optimization parameters, obtaining the target image diagnosis model, realizing deep learning-based medical image model training optimization, ensuring the adaptability of the obtained target image diagnosis model to the medical image to be diagnosed, and ensuring the accuracy of the medical diagnosis result obtained by inputting the medical image to be diagnosed into the target image diagnosis model.
[0113] Embodiment 8:
[0114] Based on Example 1, the embodiment of the present invention provides a medical image diagnosis method based on deep learning. In S4, the medical image to be diagnosed is input into a target image diagnosis model to obtain a medical diagnosis result, including:
[0115] Inputting the medical image to be diagnosed into the target image diagnosis model to obtain an image feature output result of the medical image to be diagnosed;
[0116] The image feature data result is matched with a preset diagnosis correspondence table to obtain a medical diagnosis result.
[0117] The beneficial effect of the above design scheme is: by inputting the medical image to be diagnosed into the target image diagnosis model, the image feature output result of the medical image to be diagnosed is obtained, and the medical diagnosis result is obtained based on the matching of the image feature data result with the preset diagnosis correspondence table, thereby ensuring the accuracy of the medical diagnosis result.
[0118] Embodiment 9:
[0119] Based on Example 1, the present invention provides a diagnostic system for a medical imaging diagnostic method based on deep learning, such as Figure 3 As shown, including:
[0120] The model selection module is used to pre-process the medical images to obtain the medical images to be diagnosed, and based on the image type of the medical images to be diagnosed, to match the initial image diagnosis model that meets the diagnosis requirements from the image diagnosis model database;
[0121] An image selection module, used for matching a reference medical image having a similarity greater than a preset similarity with the medical image to be diagnosed from an image database based on the image features of the medical image to be diagnosed;
[0122] A model optimization module, used to train and optimize the initial image diagnosis model using deep learning based on reference medical images to obtain a target image diagnosis model;
[0123] The medical diagnosis module is used to input the medical image to be diagnosed into the target image diagnosis model to obtain a medical diagnosis result.
[0124] In this embodiment, the preprocessing of the medical image includes image denoising, image enhancement, image smoothing and image transformation.
[0125] In this embodiment, the types of medical images to be diagnosed include vascular subtraction, magnetic resonance imaging, and ultrasound imaging.
[0126] In this embodiment, the image features of the medical image to be diagnosed include structural features, shadow features, density features, etc.
[0127] In this embodiment, the image diagnosis model collects various comprehensive diagnosis model databases formed by various types of medical images and diagnosis requirements.
[0128] In this embodiment, the image database collects various types of medical images.
[0129] The beneficial effects of the above design scheme are: by preprocessing the medical image, the medical image to be diagnosed is obtained; based on the image type of the medical image to be diagnosed, the initial image diagnosis model that meets the diagnosis requirements is matched from the image diagnosis model database to achieve the selection of the diagnosis model type, and provide a model basis for medical image diagnosis; based on the image features of the medical image to be diagnosed, a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity is matched from the image database; based on the reference medical image, the initial image diagnosis model is trained and optimized using deep learning to obtain a target image diagnosis model, and the training optimization of the medical image model based on deep learning is achieved, so as to ensure the adaptability of the target image diagnosis model to the medical image to be diagnosed, and ensure the accuracy of the medical diagnosis result obtained by inputting the medical image to be diagnosed into the target image diagnosis model.
[0130] Embodiment 10:
[0131] A storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the medical imaging diagnosis method based on deep learning as described in any one of Examples 1 to 8 are implemented.
[0132] 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 imaging diagnosis method based on deep learning, characterized in that: include: S1: preprocessing the medical image to obtain the medical image to be diagnosed, and matching the initial image diagnosis model that meets the diagnosis requirements from the image diagnosis model database based on the image type of the medical image to be diagnosed; S2: Based on the image features of the medical image to be diagnosed, matching a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity from the image database; S3: Based on the reference medical images, the initial image diagnosis model is trained and optimized by using deep learning to obtain a target image diagnosis model; S4: Inputting the medical image to be diagnosed into the target image diagnosis model to obtain a medical diagnosis result.
2. A medical imaging diagnosis method based on deep learning according to claim 1, characterized in that: In S1, preprocessing the medical image to obtain the medical image to be diagnosed includes: Based on the image type of the medical image, determining the main judgment features of the medical image, and determining specific preprocessing parameters of image denoising, image enhancement, image smoothing and image transformation based on the main judgment features; The medical image is preprocessed according to the preprocessing parameters to obtain the medical image to be diagnosed.
3. The medical imaging diagnosis method based on deep learning according to claim 1, characterized in that: In S1, based on the image type of the medical image to be diagnosed, matching an initial image diagnosis model that meets the diagnosis requirements from an image diagnosis model database includes: Based on the image type of the medical image to be diagnosed, matching and acquiring a first image diagnosis model group that satisfies the image type from an image diagnosis model database; Determine the diagnostic output result characteristics based on the diagnostic requirements of the medical image to be diagnosed, and select a second image diagnostic model group that meets the diagnostic output result characteristics from the first image diagnostic model group; Performing preliminary feature extraction on the image to be diagnosed based on a preset basic extraction method to obtain basic features of the image, obtaining training image features of each model in the second image diagnosis model group, matching the target basic features with the training image features, and obtaining a third image diagnosis model group with a matching degree greater than a preset matching degree from the second image diagnosis model group; Obtaining the latest update time of each model in the third image diagnosis model group, and selecting a fourth image diagnosis model group whose latest update time meets a preset date; Obtaining historical diagnostic data of each model in the fourth image diagnosis model group, obtaining diagnostic image features and diagnostic result features from the historical diagnostic data, determining a first weight based on the similarity between the medical image to be diagnosed and the diagnostic image features, and determining a second weight based on the diagnostic correctness of the diagnostic result features; Each model of the fourth image diagnosis model group is evaluated based on the first weight and the second weight to determine an evaluation value, and the model with the largest evaluation value is selected as the initial image diagnosis model.
4. The medical imaging diagnosis method based on deep learning according to claim 1, characterized in that: In S2, based on the image features of the medical image to be diagnosed, matching a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity from the image database includes: matching an initial medical image of the same image type as the medical image to be diagnosed from an image database; Obtaining image black shadow features, image white shadow features and image density features from image features of the medical image to be diagnosed; Based on the image black shadow feature, the image white shadow feature and the image density feature, a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity is matched from the initial medical image.
5. A medical imaging diagnosis method based on deep learning according to claim 4, characterized in that: Based on the image black shadow feature, the image white shadow feature and the image density feature, a reference medical image whose similarity with the medical image to be diagnosed is greater than a preset similarity is matched from the initial medical image, including: Based on the image type of the medical image to be diagnosed, determining feature weights for image black shadow features, image white shadow features and image density features respectively; Establishing a weighted matching rule based on the feature weights of the image black shadow feature, the image white shadow feature and the image density feature, and determining the similarity between the initial medical image and the medical image to be diagnosed according to the weighted matching rule; Matching the initial medical image with a reference medical image whose similarity is greater than a preset similarity.
6. The medical imaging diagnosis method based on deep learning according to claim 1, characterized in that: In S3, based on the reference medical images, the initial image diagnosis model is trained and optimized by using deep learning to obtain a target image diagnosis model, including: Acquire multiple identification features of the medical image to be diagnosed, mark the reference medical image based on the multiple identification features to obtain multiple marking features, and determine the marking weight of each marking feature based on the importance of each identification feature of the medical image to be diagnosed to the medical diagnosis; Obtain a first training sample based on the reference medical image, and train the initial image diagnosis model using deep learning based on the first training sample to obtain a first intermediate diagnosis model; Obtaining a second training sample based on the marking features and marking weights of the reference medical image, and training the initial image diagnosis model using deep learning based on the second training sample to obtain a second intermediate diagnosis model; Based on the model structure and model accuracy of the first intermediate diagnostic model and the model structure and model accuracy of the second intermediate diagnostic model, a functional relationship between the model structure and the model accuracy is established; Based on the functional relationship, determine the target model structure corresponding to the maximum model accuracy; The initial image diagnosis model is optimized based on the target model structure to obtain a target image diagnosis model.
7. The medical imaging diagnosis method based on deep learning according to claim 6, characterized in that: The initial image diagnosis model is optimized based on the target model structure to obtain a target image diagnosis model, including: obtaining an initial model structure of an initial imaging diagnosis model, and determining a structural difference between the initial model structure and a target model structure; Optimization parameters are determined based on the structural differences, and the initial image diagnosis model is optimized based on the optimization parameters to obtain a target image diagnosis model.
8. The medical imaging diagnosis method based on deep learning according to claim 1, characterized in that: In S4, the medical image to be diagnosed is input into the target image diagnosis model to obtain a medical diagnosis result, including: Inputting the medical image to be diagnosed into the target image diagnosis model to obtain an image feature output result of the medical image to be diagnosed; The image feature data result is matched with a preset diagnosis correspondence table to obtain a medical diagnosis result.
9. The diagnostic system of the medical imaging diagnostic method based on deep learning according to claim 1, characterized in that: include: The model selection module is used to pre-process the medical images to obtain the medical images to be diagnosed, and based on the image type of the medical images to be diagnosed, to match the initial image diagnosis model that meets the diagnosis requirements from the image diagnosis model database; An image selection module, used for matching a reference medical image having a similarity greater than a preset similarity with the medical image to be diagnosed from an image database based on the image features of the medical image to be diagnosed; A model optimization module, used to train and optimize the initial image diagnosis model using deep learning based on reference medical images to obtain a target image diagnosis model; The medical diagnosis module is used to input the medical image to be diagnosed into the target image diagnosis model to obtain a medical diagnosis result.
10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the deep learning-based medical imaging diagnosis method as described in any one of claims 1 to 8 are implemented.