Auxiliary diagnosis method and system based on clinical big data

Abstract

The invention discloses an auxiliary diagnosis method and system based on clinical big data. According to the technical scheme, a model is trained through historical clinical medical image sample data, and disease diagnosis is carried out through model assistance. The universal model and the special model are correspondingly trained for medical influences of different resolutions and different disease types, clinical medical images are judged according to the results of the universal model and the special model, medical workers are assisted in diagnosing diseases, and the disease diagnosis and treatment accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging and neural network applications, and in particular to an auxiliary diagnosis method and system based on clinical big data. Background Art

[0002] In today's era of rapid technological development, big data artificial intelligence (AI) technology is gradually becoming an important force in the medical field. Especially in the field of medical image analysis, artificial intelligence can quickly and accurately analyze a large number of medical images, such as X-rays, CT, MRI, etc., through deep learning algorithms, to help doctors detect lesions and abnormalities. At present, there are many technical solutions for medical image analysis based on artificial intelligence, and then assist in medical diagnosis and treatment, such as CN118471481B Medical multimodal auxiliary diagnosis and treatment system based on large models and artificial intelligence agents; CN114068013B A cerebral artery occlusion artificial intelligence auxiliary decision-making system; CN112786194B Medical image guidance and inspection system, method and equipment based on artificial intelligence, etc. The above patents all disclose how to train artificial intelligence through big data to achieve auxiliary diagnosis. However, in the actual application process of the above patented technologies, there are different sample numbers and types, resulting in large errors in the diagnosis results, the trained model is not accurate enough, and the same type of medical images, especially chest medical images, have various resolutions and various disease types, which further reduces the number of samples of the same resolution. Patent CN118071599B is a medical image reconstruction and recognition method and system based on machine learning, which discloses the situation of reconstructing medical images of different resolutions to high resolution through artificial intelligence and then using them as samples for training and learning. However, in the above technology, there are certain differences between the images before and after reconstruction in the corresponding learning process of artificial intelligence, and the images before and after reconstruction may give different diagnostic opinions after manual recognition and analysis, and the above patent does not provide any improvement plan. Summary of the invention

[0003] In order to address the defect that big data artificial intelligence has errors in diagnostic results after training existing medical images with different resolutions and types of images, this application proposes an auxiliary diagnosis method and system based on clinical big data.

[0004] The method and technical solution include: training a model through historical clinical medical image sample data, and using the model to assist in disease diagnosis, and the steps include: According to the resolution and diagnosis results of the acquired historical clinical medical image sample data, the historical clinical medical image sample data is set with a diagnostic disease label and a resolution label; Extracting historical clinical medical image sample data into corresponding resolution sample groups according to resolution labels, and aggregating each resolution sample group into a resolution sample group set; extracting historical clinical medical image sample data into disease sample groups according to diagnostic disease labels, and aggregating each disease sample group into a disease sample group set; For the resolution sample group set, a general resolution model applicable to each resolution of the resolution sample group set and multiple dedicated resolution models dedicated to each resolution of the resolution sample group set are trained, and the number of dedicated resolution models corresponds to the type of resolution; For the disease sample set, a general analysis model applicable to all kinds of diseases in the disease sample set and a special analysis model dedicated to each disease in the disease sample set are trained, and the number of special analysis models corresponds to the type of disease; The newly acquired clinical medical image data are respectively substituted into the general resolution model, the special resolution model of the corresponding resolution and the general analysis model. The analysis results of the general analysis model are brought into the corresponding special analysis model to determine whether the results given by each model meet the consistent standards. If so, they are output as auxiliary diagnosis results. If not, they are marked as difficult and transferred to manual processing. The processing results are used as training samples and put into the historical clinical medical image sample data to iterate the model.

[0005] The method further includes: a regular / quantitative iteration process, that is, when the newly acquired clinical medical image data and the corresponding diagnosis results are accumulated to a corresponding time / quantity, the newly acquired clinical medical image data and the corresponding diagnosis results are added to the historical clinical medical image sample data.

[0006] The method further includes that the universal resolution model also includes a reconstruction process for the sample data, and the reconstruction process includes reconstructing the low-resolution medical image into the highest resolution image in the sample data.

[0007] Further, the method also includes that the disease label includes lesion location, lesion range, disease type, disease stage and patient characteristics, and the disease label is used as a guiding condition in conjunction with the corresponding historical clinical medical image samples to participate in training, and as a constraint condition to participate in the analysis of newly acquired clinical medical image data.

[0008] Furthermore, the method also includes that the historical clinical medical image sample data is chest medical image data.

[0009] The method further includes that the general analysis model training process also includes converting the disease sample group set into a vector map, unifying the size of all the vector maps and then converting them into bitmaps before training.

[0010] Further, the method includes the process of judging whether the results given by each model meet the consistency standard, including judging that the result meets the standard when the results of the dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model are consistent; When there is inconsistency between the results of the dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model, it is determined whether the results of the dedicated resolution model and the dedicated analysis model are the same and the weight of the model with the same results prevails; if they all meet the requirements, the result is judged to meet the standard; otherwise, the result is judged to fail to meet the standard.

[0011] Further, the method also includes that the weight acquisition process of the model includes: determining the weight of each special resolution model according to the proportion of the number of images of each resolution to the total samples; determining the weight of each special analysis model according to the proportion of the number of images of each disease to the total samples; using half of the average value of the weight values ​​of each special resolution model as the weight value of the general resolution model, and using half of the average value of the weight values ​​of each special analysis model as the weight value of the general analysis model.

[0012] The present application also proposes an auxiliary diagnosis system based on clinical big data, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and the processor implements the steps in any of the above methods when executing the program.

[0013] This application trains general models and special models respectively for the medical impacts of different resolutions and different disease types, and judges clinical medical images according to the results of the general model and the special model, to assist medical staff in diagnosing diseases and improve the accuracy of disease diagnosis and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic diagram of a flow chart of an embodiment of the present application; Figure 2 This is a schematic diagram of the structure of a system for implementing an embodiment of the present application. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0016] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0017] Combination Figure 1 This embodiment proposes an auxiliary diagnosis method based on clinical big data, which trains a model through historical clinical medical image sample data and assists in disease diagnosis through the model, and the steps include: setting diagnostic disease labels and resolution labels for the historical clinical medical image sample data according to the resolution and diagnostic results of the acquired historical clinical medical image sample data; extracting the historical clinical medical image sample data into corresponding resolution sample groups according to the resolution labels, and aggregating each resolution sample group into a resolution sample group set; extracting the historical clinical medical image sample data into disease sample groups according to the diagnostic disease labels, and aggregating each disease sample group into a disease sample group set; training a general resolution model applicable to each resolution of the resolution sample group set and a resolution sample set for each resolution sample group set; A plurality of dedicated resolution models for each resolution of the group set are prepared, and the number of dedicated resolution models corresponds to the type of resolution; a general analysis model applicable to all kinds of diseases in the disease sample group set and a dedicated analysis model for each disease in the disease sample group set are trained for the disease sample group set, and the number of dedicated analysis models corresponds to the type of disease; the newly acquired clinical medical imaging data are respectively substituted into the general resolution model, the dedicated resolution model of the corresponding resolution and the general analysis model, and the analysis results of the general analysis model are brought into the corresponding dedicated analysis model to determine whether the results given by each model meet the consistent standards. If so, they are output as auxiliary diagnosis results. If not, they are marked as difficult and transferred to manual processing, and the processing results are used as training samples and put into the historical clinical medical imaging sample data to iterate the model. In this application, the existing neural network is further improved for the inaccurate defects in medical diagnosis. The dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model comprehensively diagnose the medical image to be diagnosed from four angles: the resolution image features before conversion, the general resolution image features after conversion, the matching of various disease features and the verification of the corresponding diseases. The original image before resolution conversion is diagnosed by the dedicated resolution model, and its diagnostic approach is the most direct; the general resolution model involves the most image features in diagnosis, and the image diagnosis is the most comprehensive; and the general analysis model preliminarily determines the type of disease and then further verifies whether it is accurate by the dedicated analysis model. At the same time, the dedicated analysis model can further analyze the details of the lesion. This application uses the above models for comprehensive and detailed diagnosis to improve the application of big data in clinical medicine.

[0018] Based on the above embodiment, the present embodiment is further improved and also includes a periodic / quantitative iteration process, that is, when the newly acquired clinical medical imaging data and the corresponding diagnosis results are accumulated to the corresponding time / quantity, the newly acquired clinical medical imaging data and the corresponding diagnosis results are added to the historical clinical medical imaging sample data.

[0019] Based on one or more of the above embodiments, this embodiment is further improved, and the universal resolution model also includes a reconstruction process for the sample data, and the reconstruction process includes reconstructing the low-resolution medical image into the highest resolution image in the sample data. In this application, the image reconstruction process supplements it to a higher pixel by training a reconstruction model through a neural network, or filling a pixel mode. The filling pixel is to convert the original pixel point into 4 n pixels of the same grayscale (n is a positive integer). Due to medical applications, most resolutions are exponentially multiples of 2, such as 64*64, 128*128, 256*256, etc. For image pixel values, this is an increase of 4. n Although the supplementary pixel method cannot fully reflect the image information, it can preliminarily reflect the shape outline in the image and can meet the needs of this application.

[0020] Based on one or more of the above embodiments, this embodiment is further improved, and the disease label includes lesion location, lesion range, disease type, disease stage and patient characteristics. The disease label is used as a guiding condition to cooperate with the corresponding historical clinical medical image samples to participate in the training, and as a constraint condition to participate in the analysis of newly acquired clinical medical image data. In this embodiment, through the above disease label content, the training pertinence of the model can be better improved during training, and the analysis accuracy can be improved during the analysis process. Among them, patient characteristics include indicators such as patient gender, age, height, and weight.

[0021] Based on one or more of the above embodiments, this embodiment is further improved, and the historical clinical medical image sample data is chest medical image data. This application is mainly made for chest medicine, which has a large number of medical imaging applications and a large number of diseases, to improve the diagnosis efficiency of various chest and lung diseases, improve the diagnosis and treatment efficiency of small and medium-sized hospitals, and facilitate remote diagnosis and treatment judgment.

[0022] Based on one or more of the above embodiments, this embodiment is further improved, and the general analysis model training process also includes converting the disease sample group set into a vector map, unifying the size of all the vector maps and then converting them into bitmaps for training. In this embodiment, in order not to change the proportion and position of the disease sample image and other factors, the image elements of various resolution images are maximized during the zooming and zooming process by converting the vector map. At the same time, since most neural networks can only target bitmaps, this embodiment converts the normalized image into a bitmap for training.

[0023] On the basis of one or more of the above embodiments, this embodiment is further improved, and the process of judging whether the results given by each model meet the consistent standard includes: when the results of the dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model are consistent, judging that the result meets the standard; when the results of the dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model are inconsistent, judging whether the results of the dedicated resolution model and the dedicated analysis model are the same and the weight of the model with the same result is dominant, if both meet, judging that the result meets the standard, otherwise judging that the result does not meet the standard. On the basis of one or more of the above embodiments, this embodiment is further improved, and the process of obtaining the weight of the model includes: determining the weight of each dedicated resolution model according to the proportion of the number of images of each resolution to the total samples; determining the weight of each dedicated analysis model according to the proportion of the number of images of each disease to the total samples; using half of the average value of the weight values ​​of each dedicated resolution model as the weight value of the general resolution model, and using half of the average value of the weight values ​​of each dedicated analysis model as the weight value of the general analysis model. In this embodiment, the dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model comprehensively diagnose the medical image to be diagnosed from four perspectives: the resolution image features before conversion, the general resolution image features after conversion, the matching of various disease features and the verification of the corresponding diseases. The original image before resolution conversion is diagnosed by the dedicated resolution model, which has the most direct diagnostic approach; the general resolution model involves the most image features in diagnosis, and the image diagnosis is the most comprehensive; and the general analysis model preliminarily determines the disease type and then further verifies whether it is accurate by the dedicated analysis model. At the same time, the dedicated analysis model can further analyze the details of the lesion. If there is inconsistency in the diagnosis results of the above four models, the dedicated resolution model with the most direct diagnostic approach and the dedicated analysis model with the most detailed disease judgment shall be used as the main judgment basis.

[0024] The present application also proposes an auxiliary diagnosis system based on clinical big data, including a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the program, the steps in the method described in any of the above embodiments are implemented.

[0025] Importantly, it should be noted that the construction and arrangement of the present application shown in a number of different exemplary embodiments are only exemplary. Although only a few embodiments are described in detail in this disclosure, it should be readily understood by those who refer to this disclosure that many modifications are possible (e.g., the size, scale, structure, shape and proportion of various elements, and parameter values ​​(e.g., temperature, pressure, etc.), mounting arrangements, use of materials, color, changes in orientation, etc.) without substantially departing from the novel teachings and advantages of the subject matter described in the application. For example, an element shown as integrally formed may be composed of multiple parts or elements, the position of the element may be inverted or otherwise changed, and the nature or number or position of the discrete element may be altered or changed. Therefore, all such modifications are intended to be included within the scope of the present invention. The order or sequence of any process or method steps may be changed or reordered according to an alternative embodiment. In the claims, any "device plus function" clause is intended to cover the structure described herein that performs the function, and is not only structurally equivalent but also equivalent structure. Other substitutions, modifications, changes and omissions may be made in the design, operating conditions and arrangement of the exemplary embodiments without departing from the scope of the present invention. Therefore, the invention is not limited to a specific embodiment, but extends to numerous modifications still falling within the scope of the appended claims.

[0026] Additionally, in order to provide a concise description of exemplary embodiments, all features of an actual embodiment may not be described (ie, those features that are not relevant to the best mode presently contemplated for carrying out the invention or those features that are not relevant to implementing the invention).

[0027] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. An auxiliary diagnosis method based on clinical big data, characterized in that: The model is trained by using historical clinical medical image sample data, and the model is used to assist in disease diagnosis. The steps include: According to the resolution and diagnosis results of the acquired historical clinical medical image sample data, the historical clinical medical image sample data is set with a diagnostic disease label and a resolution label; Extracting historical clinical medical image sample data into corresponding resolution sample groups according to resolution labels, and aggregating each resolution sample group into a resolution sample group set; extracting historical clinical medical image sample data into disease sample groups according to diagnostic disease labels, and aggregating each disease sample group into a disease sample group set; For the resolution sample group set, a general resolution model applicable to each resolution of the resolution sample group set and multiple dedicated resolution models dedicated to each resolution of the resolution sample group set are trained, and the number of dedicated resolution models corresponds to the type of resolution; For the disease sample set, a general analysis model applicable to all kinds of diseases in the disease sample set and a special analysis model dedicated to each disease in the disease sample set are trained, and the number of special analysis models corresponds to the type of disease; The newly acquired clinical medical image data are respectively substituted into the general resolution model, the special resolution model of the corresponding resolution and the general analysis model. The analysis results of the general analysis model are brought into the corresponding special analysis model to determine whether the results given by each model meet the consistent standards. If so, they are output as auxiliary diagnosis results. If not, they are marked as difficult and transferred to manual processing. The processing results are used as training samples and put into the historical clinical medical image sample data to iterate the model.

2. The method according to claim 1, characterized in that It also includes a regular / quantitative iteration process, that is, when the newly acquired clinical medical imaging data and the corresponding diagnosis results are accumulated to the corresponding time / quantity, the newly acquired clinical medical imaging data and the corresponding diagnosis results are added to the historical clinical medical imaging sample data.

3. The method according to claim 1, characterized in that The universal resolution model also includes a reconstruction process for the sample data, wherein the reconstruction process includes reconstructing the low-resolution medical image into the highest-resolution image in the sample data.

4. The method according to claim 1, characterized in that: The disease labels include lesion location, lesion range, disease type, disease stage and patient characteristics. The disease labels are used as guiding conditions in conjunction with corresponding historical clinical medical image samples to participate in training, and as constraints to participate in the analysis of newly acquired clinical medical image data.

5. The method according to claim 1, characterized in that The historical clinical medical image sample data is chest medical image data.

6. The method according to claim 1, characterized in that The general analysis model training process also includes converting the disease sample group set into a vector map, unifying the size of all the vector maps and then converting them into bitmaps for training.

7. The method according to claim 1, characterized in that The process of judging whether the results given by each model meet the consistency standard includes: when the results of the dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model are consistent, judging that the result meets the standard; When there is inconsistency between the results of the dedicated resolution model, the general resolution model, the general analysis model and the dedicated analysis model, it is determined whether the results of the dedicated resolution model and the dedicated analysis model are the same and the weight of the model with the same results prevails; if they all meet the requirements, the result is judged to meet the standard; otherwise, the result is judged to fail to meet the standard.

8. The method according to claim 7, characterized in that The weight acquisition process of the model includes: determining the weight of each special resolution model according to the ratio of the number of images of each resolution to the total samples; determining the weight of each special analysis model according to the ratio of the number of images of each disease to the total samples; using half of the average value of the weight values ​​of each special resolution model as the weight value of the general resolution model, and using half of the average value of the weight values ​​of each special analysis model as the weight value of the general analysis model.

9. An auxiliary diagnosis system based on clinical big data, comprising a memory and a processor, wherein the memory stores a computer program that can be run on the processor, characterized in that: When the processor executes the program, the steps in the method according to any one of claims 1 to 8 are implemented.

Images