Medical institution-oriented AI data governance method
By evaluating and filtering the rationality of medical text and image data, and optimizing the AI model by combining historical diagnostic big data, the problem of insufficient training quality of medical diagnostic models in existing technologies has been solved, and the accuracy and applicability of interaction have been improved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 福鑫数科(杭州)人工智能有限公司
- Filing Date
- 2026-01-23
- Publication Date
- 2026-05-05
AI Technical Summary
Existing AI data governance methods are insufficient for assessing the reasonableness of medical text and image data in medical institutions, resulting in inadequate training quality of medical diagnostic models and affecting the accuracy of interactions.
By performing symptom diagnosis matching and indicator diagnosis matching analysis on medical text data, and combining it with the similarity assessment of image data, highly reasonable data is selected for model training, and a dynamic governance mechanism is constructed to optimize the diagnostic model.
This improves the accuracy of medical diagnostic models in interacting with medical text and image data, ensures the quality of training data, reduces noise interference, and enhances the accuracy and applicability of diagnostic models.
Smart Images
Figure CN121983283A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of smart healthcare and involves artificial intelligence technology, specifically an AI data governance method for medical institutions. Background Technology
[0002] Existing AI data governance methods have the following shortcomings when governing medical text data and medical image data in medical institutions: 1. Existing AI data governance methods, when governing medical text data, can usually only perform formatted comparison and missing value supplementation. They are unable to combine historical diagnostic big data from medical institutions to evaluate the rationality of clinical diagnoses in medical text data. They cannot use the evaluation to govern medical texts with unreasonable diagnoses using AI models, nor can they use medical texts with reasonable diagnoses derived by AI models to train medical diagnostic models. As a result, it is difficult to guarantee the text training quality of medical diagnostic models, leading to insufficient interactive accuracy when medical diagnostic models process medical text data. 2. Existing AI data governance methods, when governing medical text data, can usually only perform formatted comparison and clarity restoration of medical image data. They are unable to combine historical image diagnosis big data from medical institutions to evaluate the rationality of image diagnoses in medical image data. They cannot use AI models to govern medical images with unreasonable diagnoses based on the evaluation results, nor can they use the medical images with reasonable diagnoses output by the AI model to train the medical diagnostic model. This makes it difficult to guarantee the quality of graphic training for the medical diagnostic model, thus affecting the quality of interaction with medical image data.
[0003] To this end, we propose an AI data governance approach for medical institutions. Summary of the Invention
[0004] In view of the shortcomings of existing technologies, the purpose of this invention is to provide an AI data governance method for medical institutions, which aims to improve the quality of AI governance of medical data.
[0005] To achieve the above objectives, the present invention adopts the following technical solution: an AI data governance method for medical institutions, comprising the following steps: Step S1: Acquire medical text data and medical image data, divide the medical text data into independent medical analysis texts, perform diagnostic rationality analysis on the medical analysis texts, screen the medical analysis texts according to the analysis results, obtain the initial screening data of medical texts, and define the initial screening data of medical texts and medical image data as data collected by medical institutions. Step S2: Perform diagnostic matching analysis on medical analysis images based on data collected by medical institutions, obtain the image diagnostic rationality index corresponding to the medical analysis images based on the analysis results, and screen the medical analysis images based on the image diagnostic rationality index to obtain the initial screening data of medical images. Step S3: Perform medical data processing based on the initial screening data of medical images and the data collected by medical institutions to obtain qualified data.
[0006] Furthermore, step S1 also includes the following steps: Step S11: Acquire the medical institutions that need to undergo AI governance of medical data, and randomly select one target medical institution from the multiple acquired medical institutions; Step S12: Obtain the medical text generated by the target medical institution at the current moment to obtain medical text data; obtain the medical images generated by the target medical institution at the current moment to obtain medical image data. Step S13: Conduct a governance needs analysis on the medical text data, and obtain the initial screening data of medical texts based on the analysis results; Step S14: Define the initial screening data of medical texts and medical image data as data collected by medical institutions.
[0007] Furthermore, step S13 also includes the following steps: Step S131: Divide the medical text data into multiple medical analysis texts, and arbitrarily select one sample analysis text from the multiple medical analysis texts obtained; Step S132: Perform symptom diagnosis matching analysis on the sample analysis text, and obtain the symptom diagnosis matching degree corresponding to the sample analysis text based on the analysis results; Step S133: Perform indicator diagnostic matching analysis on the sample analysis text, and obtain the indicator diagnostic matching degree corresponding to the sample analysis text based on the analysis results; Step S134: Calculate the text diagnosis rationality index corresponding to the sample analysis text by matching the symptom diagnosis matching degree and the indicator diagnosis matching degree with the sample analysis text. The text diagnostic rationality index corresponding to the sample analysis text is calculated using the following formula: ; Where Whz is the text diagnosis rationality index corresponding to the sample analysis text, Zbz is the indicator diagnosis matching degree corresponding to the sample analysis text, Zpz is the symptom diagnosis matching degree corresponding to the sample analysis text, w1 is the first matching degree weight, and w2 is the second matching degree weight. Step S135: Obtain the text diagnosis rationality index corresponding to each medical analysis text.
[0008] Furthermore, step S132 also includes the following steps: The clinical diagnosis is obtained from the sample analysis text to obtain the target clinical diagnosis; The clinical symptoms of patients involved in the sample analysis text were obtained, resulting in the clinical symptoms of multiple patients; AI analysis tools are used to acquire typical symptoms involved in the target clinical diagnosis, resulting in multiple typical symptoms for diagnosis. The clinical symptoms of multiple patients and the same symptoms among the multiple typical symptoms for diagnosis are acquired to obtain multiple common symptoms for diagnosis. The acquired common symptoms for diagnosis are marked as Z1 common clinical symptoms to Za common clinical symptoms. By using AI analysis tools, we can obtain historical patients with the target clinical diagnosis who have been treated at the target medical institution. We can obtain multiple historical target patients, count the number of historical target patients, and obtain the cumulative number of historical target diagnoses. Among the acquired historical target patients, the number of historical target patients with Z1 common clinical symptoms was counted to obtain the number of Z1 symptom cases. The ratio of the number of Z1 symptom cases to the cumulative number of historical target diagnoses was calculated to obtain the historical occurrence percentage of Z1 symptoms. The number of historical target patients with Z2 common clinical symptoms was counted to obtain the number of Z2 symptom cases. The ratio of the number of Z2 symptom cases to the cumulative number of historical target diagnoses was calculated to obtain the historical occurrence percentage of Z2 symptoms. Similarly, the number of historical target patients with Za common clinical symptoms was counted to obtain the number of Za symptom cases. The ratio of the number of Za symptom cases to the cumulative number of historical target diagnoses was calculated to obtain the historical occurrence percentage of Za symptoms. Among the multiple historical target patients obtained, the number of symptom cases corresponding to each diagnostic typical symptom is obtained, and the ratio of the obtained number of symptom cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage corresponding to each typical symptom. The multiple historical occurrence percentages are summed to obtain the cumulative occurrence percentage of typical symptoms. The symptom diagnosis matching degree corresponding to the sample analysis text is obtained by calculating the percentage of historical occurrence of Z1 symptoms, the percentage of historical occurrence of Za symptoms, and the cumulative percentage of occurrence of typical symptoms. The symptom diagnosis matching degree corresponding to the sample analysis text is calculated using the following formula: ; Where Zpz represents the symptom diagnosis matching degree corresponding to the sample analysis text, Zlzi represents the historical occurrence percentage of Zi symptoms, and Zlh represents the cumulative occurrence percentage of typical symptoms.
[0009] Furthermore, step S133 also includes the following steps: The clinical diagnosis is obtained by using a text recognition algorithm to analyze the sample text. Abnormal medical indicators involved in the sample analysis text were obtained, resulting in multiple abnormal indicators for patients. AI analysis tools are used to acquire typical medical indicators involved in the target clinical diagnosis, resulting in multiple typical diagnostic indicators. Abnormal indicators of multiple patients and the same medical conditions among the multiple typical diagnostic indicators are acquired to obtain multiple common diagnostic medical indicators. The acquired common diagnostic symptoms are marked as Z1 common medical indicators to Zb common medical indicators. Multiple historical target patients are acquired and their numbers are statistically analyzed to obtain the cumulative number of historical target diagnoses. Among the acquired historical target patients, the number of historical target patients with the Z1 common medical indicator is statistically analyzed to obtain the number of Z1 indicator cases. The ratio of the number of Z1 indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage of the Z1 indicator. The number of historical target patients with the Z2 common medical indicator is statistically analyzed to obtain the number of Z2 indicator cases. The ratio of the number of Z2 indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage of the Z2 indicator. This process is repeated for historical target patients with the Zb common medical indicator. The number of Zb indicator cases is statistically analyzed to obtain the number of Zb indicator cases. The ratio of the number of Zb indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage of the Zb indicator. Among the multiple historical target patients obtained, the number of cases corresponding to each diagnostic typical indicator is obtained, and the ratio of the obtained number of indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage corresponding to each diagnostic typical indicator. The multiple historical occurrence percentages are summed to obtain the cumulative occurrence percentage of typical indicators. The historical occurrence percentage of the Z1 indicator, the historical occurrence percentage of the Zb indicator, and the cumulative occurrence percentage of typical indicators are used to calculate the indicator diagnostic matching degree corresponding to the sample analysis text. The diagnostic matching degree of the indicators corresponding to the sample analysis text is calculated using the following formula: ; Where Zbz represents the diagnostic matching degree of the indicator corresponding to the sample analysis text, Zbzi represents the historical occurrence percentage of the Zi indicator, and Zbh represents the cumulative occurrence percentage of the typical indicator.
[0010] Furthermore, step S2 also includes the following steps: Step S21: Obtain data collected by medical institutions, obtain medical image data based on the data collected by medical institutions, split the medical image data into several medical analysis images, and arbitrarily select one sample analysis image from the multiple medical analysis images obtained; Step S22: Perform diagnostic matching degree analysis on the sample analysis images, and obtain the image diagnosis rationality index corresponding to the sample analysis images based on the analysis results; Step S23: Obtain the image diagnosis rationality index corresponding to each medical analysis image; Step S24: Obtain the preset range of the reasonable index for image diagnosis. If the reasonable index for image diagnosis is within the preset range, the corresponding medical analysis image is screened as a qualified medical image. If the reasonable index for image diagnosis is not within the preset range, the corresponding medical analysis image is screened as a medical image to be treated, thus obtaining the initial screening data for medical images.
[0011] Furthermore, step S22 also includes the following steps: Step S221: Obtain the medical diagnosis result corresponding to the sample image analysis to obtain the sample image diagnosis; Step S222: Acquire historical medical images for which the medical diagnosis result is a sample image diagnosis, obtain multiple historical analysis images, and arbitrarily select one sample historical image from the multiple acquired historical analysis images; Step S223: If the sample image is diagnosed as lumbar disc herniation, perform a similarity analysis between the sample analysis image and the sample historical image, and obtain the medical image similarity between the sample analysis image and the sample historical image based on the analysis results. Step S223 further includes the following steps: Adjust the sample analysis image and the sample historical image to the same image scaling ratio, and use the geometric center corresponding to the sample analysis image as the origin to create a plane rectangular coordinate system, thus obtaining the image plane rectangular coordinate system; The geometric center point of the vertebral body at the upper end of the protruding intervertebral disc in the sample analysis image is set as the first vertebral body feature point for analysis. The geometric center point of the vertebral body at the lower end of the protruding intervertebral disc in the sample analysis image is set as the second vertebral body feature point for analysis. The geometric center point of the vertebral body at the upper end of the protruding intervertebral disc in the sample historical image is set as the first historical vertebral body feature point. The geometric center point of the vertebral body at the lower end of the protruding intervertebral disc in the sample historical image is set as the second historical vertebral body feature point. The line connecting the first vertebral body feature point and the second vertebral body feature point is set as the vertebral body connection line for analysis. The line connecting the first historical vertebral body feature point and the second historical vertebral body feature point is set as the historical vertebral body connection line.
[0012] Furthermore, step S223 also includes the following steps: In the Cartesian coordinate system of the image plane, the sample historical image is covered by the sample analysis image. If the length of the analysis vertebral body connection is greater than or equal to that of the historical vertebral body connection, the analysis vertebral body connection will completely cover the historical vertebral body connection. If the length of the sample vertebral body connection is less than that of the historical vertebral body connection, the historical vertebral body connection will completely cover the analysis vertebral body connection. The lumbar intervertebral disc region is obtained from the sample analysis image to obtain the first lumbar intervertebral disc region, and the lumbar intervertebral disc region is obtained from the sample historical image to obtain the second lumbar intervertebral disc region. The number of pixels within the contour of the first lumbar intervertebral disc region is counted to obtain the total number of pixels within the first contour. The number of pixels at the contour boundary of the first lumbar intervertebral disc region is counted to obtain the total number of pixels at the contour boundary. The roundness of the first intervertebral disc is obtained by calculating the total number of pixels within the first contour and the total number of pixels at the boundary of the first contour. The number of pixels within the contour of the second lumbar intervertebral disc region is counted to obtain the total number of pixels within the second contour. The number of pixels at the contour boundary of the second lumbar intervertebral disc region is counted to obtain the total number of pixels at the contour boundary. The roundness of the second intervertebral disc is obtained by calculating the total number of pixels within the second contour and the total number of pixels at the boundary of the second contour.
[0013] Furthermore, step S223 also includes the following steps: Calculate the difference between the roundness of the first intervertebral disc and the roundness of the second intervertebral disc, and calculate the ratio of the obtained difference to the roundness of the second intervertebral disc to obtain the lumbar disc roundness deviation in the imaging. The overlapping area between the first and second lumbar intervertebral disc regions is obtained to obtain the third lumbar intervertebral disc region. The number of pixels in the first, second, and third lumbar intervertebral disc regions is counted to obtain the pixel count values of the first region, the second region, and the third region. The overlap rate of the intervertebral disc region in the image is obtained by calculating the pixel count values of the first region, the second region, and the third region. The overlap rate of the lumbar intervertebral disc region in imaging is calculated using the following formula: ; Where Ych is the overlap rate of the lumbar intervertebral disc region in the image, Qsz1 is the number of pixels in the first region, Qsz2 is the number of pixels in the second region, and Qsz3 is the number of pixels in the third region. The medical image similarity between the sample analysis image and the sample historical image is obtained by calculating the overlap rate of the lumbar intervertebral disc region and the lumbar intervertebral disc roundness deviation. The medical image similarity between the sample analysis images and the sample historical images is calculated using the following formula: ; Where Yxs is the medical image similarity between the sample analysis image and the sample historical image, Ych is the overlap rate of the lumbar intervertebral disc region in the image, and Ydc is the lumbar intervertebral disc roundness deviation in the image. The medical image similarity between the sample analysis image and each historical analysis image is obtained separately, resulting in multiple medical image similarities. The medical image similarity with the largest value is set as the reasonable index for image diagnosis corresponding to the sample analysis image.
[0014] Furthermore, step S3 also includes the following steps: Acquire data collected by medical institutions, and obtain preliminary screening data for medical texts based on the data collected by medical institutions; Multiple medical texts to be processed are obtained based on the initial screening data of medical texts. These texts are then fed back to the text upload terminal, where medical staff correct them to obtain the texts to be judged. A text processing judgment model is then created. If the text processing judgment model screens the texts to be judged as qualified medical texts, the processing of the texts to be processed is complete. If the text processing judgment model screens the texts to be judged as unprocessable medical texts, the processing of the texts to be judged is repeated until the text processing judgment model screens the texts to be judged as qualified medical texts. Multiple medical images to be treated are obtained based on the initial screening data of medical images. These images are then fed back to the image upload terminal, where medical staff correct them to obtain images to be judged. An image treatment judgment model is then created. If the image treatment judgment model screens the images to be judged as qualified images, the treatment of the images to be treated is completed. If the image treatment judgment model screens the images to be judged as images to be treated, the treatment of the images to be treated is repeated until the image treatment judgment model screens the images to be judged as qualified images. Qualified medical images and qualified medical texts are defined as qualified data for governance.
[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention evaluates the rationality of clinical diagnoses in medical text data by combining historical diagnostic big data from medical institutions. Based on the evaluation, it uses an AI model to manage medical texts with unreasonable diagnoses. The reasonable medical texts derived by the AI model are then used to train the medical diagnostic model, which ensures the quality of text training for the medical diagnostic model and improves the accuracy of the medical diagnostic model's interaction with medical text data. 2. This invention evaluates the rationality of image diagnoses in medical image data by combining historical image diagnostic big data from medical institutions. Based on the evaluation results, AI models are used to manage medical images with unreasonable diagnoses. The medical images with reasonable diagnoses output by the AI model are used to train the medical diagnostic model, which can ensure the quality of graphic training of the medical diagnostic model and thus improve the interaction quality of medical image data. Attached Figure Description
[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.
[0017] Figure 1 This is a diagram illustrating the implementation steps of the present invention; Figure 2 This is a schematic diagram of the vertebral body connection in this invention; Figure 3 This is a schematic diagram of the third lumbar intervertebral disc region of the present invention. Detailed Implementation
[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0019] Example 1 Please see Figure 1 This invention provides a technical solution: an AI data governance method for medical institutions, comprising the following steps: Step S1: Acquire medical text data and medical image data, divide the medical text data into independent medical analysis texts, perform diagnostic rationality analysis on the medical analysis texts, screen the medical analysis texts according to the analysis results, obtain the initial screening data of medical texts, and define the initial screening data of medical texts and medical image data as data collected by medical institutions. Step S1 further includes the following steps: The system acquires data from medical institutions that require AI-based medical data governance, and then randomly selects one target medical institution from among the acquired institutions. The medical texts generated by the target medical institution at the current moment are acquired to obtain medical text data, and the medical images generated by the target medical institution at the current moment are acquired to obtain medical image data. It should be noted here that: In this application, the medical texts involved herein include, but are not limited to, electronic medical records, outpatient records, and surgical records; and the medical images involved herein include, but are not limited to, CT images, DR images, and MRC images. In this application, the medical text data and medical image data involved herein may come from different business systems, including but not limited to HIS (Hospital Information System), EMR (Electronic Medical Record System), PACS (Picture Archiving and Communication System), and LIS (Laboratory Information System).
[0020] A governance needs analysis was conducted on medical text data, and preliminary screening data of medical texts was obtained based on the analysis results. Specifically as follows: The medical text data is divided into multiple medical analysis texts, and one sample analysis text is randomly selected from the multiple medical analysis texts. It should be noted here that: In this application, the medical analysis text referred to herein specifically refers to a single patient's medical record, a single patient's outpatient record, or a single surgical record; It should be noted here that: The basic information fields in the sample analysis text are collected. If any basic information field is missing, the sample analysis text is fed back to medical staff or patients so that they can supplement and correct the basic information fields. It should be noted here that: In this application, the basic information fields referred to herein are specifically the patient's personal information in the medical analysis text, including but not limited to name, age, and gender; In this application, the medical analysis text referred to herein includes textual content related to the patient's condition output by medical personnel or the patient's own description of the symptoms. The medical information fields referred to herein include, but are not limited to, the patient's chief complaint, clinical diagnosis, and treatment recommendations.
[0021] Perform symptom diagnosis matching analysis on the sample analysis text, and obtain the symptom diagnosis matching degree corresponding to the sample analysis text based on the analysis results; Specifically as follows: The clinical diagnosis is obtained from the sample analysis text to obtain the target clinical diagnosis; The clinical symptoms of patients involved in the sample analysis text were obtained, resulting in the clinical symptoms of multiple patients; AI analysis tools are used to acquire typical symptoms involved in the target clinical diagnosis, resulting in multiple typical symptoms for diagnosis. The clinical symptoms of multiple patients and the same symptoms among the multiple typical symptoms for diagnosis are acquired to obtain multiple common symptoms for diagnosis. The acquired common symptoms for diagnosis are marked as Z1 common clinical symptoms to Za common clinical symptoms. It should be noted here that: In the application, the AI analysis tool mentioned here is an AI medical diagnostic model created by the target medical institution. The AI medical diagnostic model mentioned here is a big data question-answering model trained from qualified medical text data and qualified medical image data. In this application, the diagnostic typical symptoms referred to herein are characteristic clinical manifestations that appear during the development of a disease and are highly suggestive of the diagnosis. If the target clinical diagnosis is myocarditis, the diagnostic typical symptoms include, but are not limited to, chest tightness, chest pain, and shortness of breath. In this application, Z1 represents the number of shared clinical symptoms, up to Za represents the number of shared clinical diagnoses, where a is an integer greater than 0.
[0022] By using AI analysis tools, we can obtain historical patients with the target clinical diagnosis who have been treated at the target medical institution. We can obtain multiple historical target patients, count the number of historical target patients, and obtain the cumulative number of historical target diagnoses. Among the acquired historical target patients, the number of historical target patients with Z1 common clinical symptoms was counted to obtain the number of Z1 symptom cases. The ratio of the number of Z1 symptom cases to the cumulative number of historical target diagnoses was calculated to obtain the historical occurrence percentage of Z1 symptoms. The number of historical target patients with Z2 common clinical symptoms was counted to obtain the number of Z2 symptom cases. The ratio of the number of Z2 symptom cases to the cumulative number of historical target diagnoses was calculated to obtain the historical occurrence percentage of Z2 symptoms. Similarly, the number of historical target patients with Za common clinical symptoms was counted to obtain the number of Za symptom cases. The ratio of the number of Za symptom cases to the cumulative number of historical target diagnoses was calculated to obtain the historical occurrence percentage of Za symptoms. Among the multiple historical target patients obtained, the number of symptom cases corresponding to each diagnostic typical symptom is obtained, and the ratio of the obtained number of symptom cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage corresponding to each typical symptom. The multiple historical occurrence percentages are summed to obtain the cumulative occurrence percentage of typical symptoms. The symptom diagnosis matching degree corresponding to the sample analysis text is obtained by calculating the percentage of historical occurrence of Z1 symptoms, the percentage of historical occurrence of Za symptoms, and the cumulative percentage of occurrence of typical symptoms. The symptom diagnosis matching degree corresponding to the sample analysis text is calculated using the following formula: ; Where Zpz represents the symptom diagnosis matching degree corresponding to the sample analysis text, Zlzi represents the historical occurrence percentage of Zi symptoms, and Zlh represents the cumulative occurrence percentage of typical symptoms. It should be noted here that: In this application, the percentage of Zi symptoms in history can be any one of the percentages of Z1 symptoms in history to Za symptoms in history. Perform indicator diagnostic matching analysis on the sample analysis text, and obtain the indicator diagnostic matching degree corresponding to the sample analysis text based on the analysis results; Specifically as follows: The clinical diagnosis is obtained by using a text recognition algorithm to analyze the sample text. Abnormal medical indicators involved in the sample analysis text were obtained, resulting in multiple abnormal indicators for patients. AI analysis tools are used to acquire typical medical indicators involved in the target clinical diagnosis, resulting in multiple typical diagnostic indicators. Abnormal indicators of multiple patients and the same medical conditions among the multiple typical diagnostic indicators are acquired to obtain multiple common diagnostic medical indicators. The acquired common diagnostic symptoms are marked as Z1 common medical indicators to Zb common medical indicators. It should be noted here that: In the application, the AI analysis tool mentioned here is an AI medical diagnostic model created by the target medical institution. The AI medical diagnostic model mentioned here is a big data question-answering model trained from qualified medical text data and qualified medical image data. In this application, the diagnostic typical indicators referred to herein are medical test indicators that are characteristic and highly suggestive of disease diagnosis during the development of a disease. If the target clinical diagnosis is myocarditis, the diagnostic typical indicators include, but are not limited to, troponin, creatine kinase, and C-reactive protein. In this application, Z1, Zb, and Zb are the numbers corresponding to the shared medical indicators, from 1 to 2 to 3 to b.
[0023] Multiple historical target patients are acquired and their numbers are statistically analyzed to obtain the cumulative number of historical target diagnoses. Among the acquired historical target patients, the number of historical target patients with the Z1 common medical indicator is statistically analyzed to obtain the number of Z1 indicator cases. The ratio of the number of Z1 indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage of the Z1 indicator. The number of historical target patients with the Z2 common medical indicator is statistically analyzed to obtain the number of Z2 indicator cases. The ratio of the number of Z2 indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage of the Z2 indicator. This process is repeated for historical target patients with the Zb common medical indicator. The number of Zb indicator cases is statistically analyzed to obtain the number of Zb indicator cases. The ratio of the number of Zb indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage of the Zb indicator. Among the multiple historical target patients obtained, the number of cases corresponding to each diagnostic typical indicator is obtained, and the ratio of the obtained number of indicator cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage corresponding to each diagnostic typical indicator. The multiple historical occurrence percentages are summed to obtain the cumulative occurrence percentage of typical indicators. The historical occurrence percentage of the Z1 indicator, the historical occurrence percentage of the Zb indicator, and the cumulative occurrence percentage of typical indicators are used to calculate the indicator diagnostic matching degree corresponding to the sample analysis text. The diagnostic matching degree of the indicators corresponding to the sample analysis text is calculated using the following formula: ; Where Zbz represents the diagnostic matching degree of the indicator corresponding to the sample analysis text, Zbzi represents the historical occurrence percentage of the Zi indicator, and Zbh represents the cumulative occurrence percentage of the typical indicator. It should be noted here that: In this application, the historical occurrence percentage of the Zi indicator can be any one of the historical occurrence percentages of the Z1 indicator to the Zb indicator. The text diagnosis rationality index corresponding to the sample analysis text is obtained by calculating the symptom diagnosis matching degree and the indicator diagnosis matching degree of the sample analysis text. The text diagnostic rationality index corresponding to the sample analysis text is calculated using the following formula: ; Where Whz is the text diagnosis rationality index corresponding to the sample analysis text, Zbz is the indicator diagnosis matching degree corresponding to the sample analysis text, Zpz is the symptom diagnosis matching degree corresponding to the sample analysis text, w1 is the first matching degree weight, and w2 is the second matching degree weight. It should be noted here that: The sum of the first matching degree weight and the second matching degree weight involved here is 1, and the specific values of the first matching degree weight and the second matching degree weight need to be determined based on the target clinical diagnosis.
[0024] Repeat the process of obtaining the text diagnostic rationality index corresponding to the sample analysis text, and obtain the text diagnostic rationality index corresponding to each medical analysis text respectively; Obtain the preset range of the reasonable index for text diagnosis. If the reasonable index of text diagnosis is within the preset range, the corresponding medical analysis text is screened as qualified medical text. If the reasonable index of text diagnosis is not within the preset range, the corresponding medical analysis text is screened as medical text to be treated, thus obtaining the initial screening data of medical text. It should be noted here that: The system obtains historical qualified medical texts that have been screened by AI analysis tools, obtains the text diagnosis rationality index corresponding to each historical qualified medical text, compares the values of multiple text diagnosis rationality indices, sets the text diagnosis rationality index with the largest value as the first feature text rationality index, sets the text diagnosis rationality index with the smallest value as the second feature text rationality index, and sets the numerical range composed of the first feature text rationality index and the second feature text rationality index as the preset range of text diagnosis rationality index. In this application, a qualified medical text includes cases where the text diagnostic reasonableness index is within the preset range boundary of the text diagnostic reasonableness index.
[0025] Medical text screening data and medical image data are defined as data collected by medical institutions. It should be noted here that: Step S1 above assesses the rationality of clinical diagnoses in medical text data by combining historical diagnostic big data from medical institutions. Based on the assessment, AI models are used to manage medical texts with unreasonable diagnoses. The reasonable medical texts identified by the AI model are used to train the medical diagnostic model. This can build an evaluation benchmark based on massive historical diagnostic data, effectively identify and filter abnormal diagnostic texts with logical contradictions or insufficient clinical evidence, and ensure the quality of training data from the source. By continuously optimizing the diagnosis of unreasonable text through a dynamic governance mechanism, noise interference in model training is reduced, and a closed-loop optimization system of "evaluation-governance-iteration" is formed. The final adopted diagnostic medical text has stronger clinical representativeness and diagnostic consistency, which can significantly improve the medical diagnostic model's ability to learn representations of complex diseases and its ability to generalize across scenarios, ultimately achieving a dual improvement in diagnostic accuracy and clinical applicability.
[0026] Step S2: Perform diagnostic matching analysis on medical analysis images based on data collected by medical institutions, obtain the image diagnostic rationality index corresponding to the medical analysis images based on the analysis results, and screen the medical analysis images based on the image diagnostic rationality index to obtain the initial screening data of medical images. Step S2 further includes the following steps: Acquire data collected by medical institutions, obtain medical imaging data based on the data collected by medical institutions, split the medical imaging data into several medical analysis images, and arbitrarily select one sample image from the multiple acquired medical analysis images for analysis. Perform diagnostic matching degree analysis on the sample analysis images, and obtain the image diagnosis rationality index corresponding to the sample analysis images based on the analysis results; Specifically as follows: The medical diagnostic results corresponding to the sample images are obtained to obtain the sample image diagnosis; It should be noted here that: In this application, the sample imaging diagnosis involved herein can be lumbar disc herniation, lumbar disc bulging, and lumbar disc prolapse.
[0027] The historical medical images for which the medical diagnosis result is a sample image diagnosis are acquired, resulting in multiple historical analysis images. Then, one sample historical image is randomly selected from the multiple acquired historical analysis images. It should be noted here that: In this application, the historical analysis images involved are historical medical images stored by the target medical institution, and the medical diagnosis reflected by the historical analysis images and the sample analysis images is the same medical diagnosis, and the image scanning areas corresponding to the historical analysis images and the sample analysis images are the same.
[0028] In this application, the image diagnostic rationality index corresponding to the historical analysis images involved herein is qualified.
[0029] If the sample image is diagnosed as lumbar disc herniation, a similarity analysis is performed between the sample analysis image and the sample historical image, and the medical image similarity between the sample analysis image and the sample historical image is obtained based on the analysis results. Specifically as follows: Adjust the sample analysis image and the sample historical image to the same image scaling ratio, and use the geometric center corresponding to the sample analysis image as the origin to create a plane rectangular coordinate system, thus obtaining the image plane rectangular coordinate system; Please see Figure 2The geometric center point of the vertebral body at the upper end of the protruding intervertebral disc in the sample analysis image is set as the first vertebral body feature point, the geometric center point of the vertebral body at the lower end of the protruding intervertebral disc in the sample analysis image is set as the second vertebral body feature point, the geometric center point of the vertebral body at the upper end of the protruding intervertebral disc in the sample historical image is set as the first historical vertebral body feature point, the geometric center point of the vertebral body at the lower end of the protruding intervertebral disc in the sample historical image is set as the second historical vertebral body feature point, the line connecting the first vertebral body feature point and the second vertebral body feature point is set as the vertebral body connection line, and the line connecting the first historical vertebral body feature point and the second historical vertebral body feature point is set as the historical vertebral body connection line. It should be noted here that: In this application, only one lumbar disc herniation is found in the sample analysis images and the sample historical images, and the location of the herniation is the same.
[0030] In the Cartesian coordinate system of the image plane, the sample historical image is covered by the sample analysis image. If the length of the analysis vertebral body connection is greater than or equal to that of the historical vertebral body connection, the analysis vertebral body connection will completely cover the historical vertebral body connection. If the length of the sample vertebral body connection is less than that of the historical vertebral body connection, the historical vertebral body connection will completely cover the analysis vertebral body connection. The lumbar intervertebral disc region is obtained from the sample analysis image to obtain the first lumbar intervertebral disc region, and the lumbar intervertebral disc region is obtained from the sample historical image to obtain the second lumbar intervertebral disc region. The number of pixels within the contour of the first lumbar intervertebral disc region is counted to obtain the total number of pixels within the first contour. The number of pixels at the contour boundary of the first lumbar intervertebral disc region is counted to obtain the total number of pixels at the contour boundary. The roundness of the first intervertebral disc is obtained by calculating the total number of pixels within the first contour and the total number of pixels at the boundary of the first contour. The specific formula for calculating the roundness of the first intervertebral disc is as follows: ; Where Yzd1 is the roundness of the first intervertebral disc, Lkn is the total number of pixels within the first contour, and Lkb is the total number of pixels at the boundary of the first contour. Repeat the process of acquiring the roundness of the first intervertebral disc, acquire the roundness of the second intervertebral disc numerically, calculate the difference between the roundness of the first and second intervertebral discs, and calculate the ratio of the obtained difference to the roundness of the second intervertebral disc to obtain the lumbar disc roundness deviation in the imaging. Please see Figure 3 The overlapping area between the first and second lumbar intervertebral disc regions is obtained to obtain the third lumbar intervertebral disc region. The number of pixels in the first, second, and third lumbar intervertebral disc regions is counted to obtain the pixel count values of the first region, the second region, and the third region. The overlap rate of the intervertebral disc region in the image is obtained by calculating the pixel count values of the first region, the second region, and the third region. The overlap rate of the lumbar intervertebral disc region in imaging is calculated using the following formula: ; Where Ych is the overlap rate of the lumbar intervertebral disc region in the image, Qsz1 is the number of pixels in the first region, Qsz2 is the number of pixels in the second region, and Qsz3 is the number of pixels in the third region. The medical image similarity between the sample analysis image and the sample historical image is obtained by calculating the overlap rate of the lumbar intervertebral disc region and the lumbar intervertebral disc roundness deviation. The medical image similarity between the sample analysis images and the sample historical images is calculated using the following formula: ; Where Yxs is the medical image similarity between the sample analysis image and the sample historical image, Ych is the overlap rate of the lumbar intervertebral disc region in the image, and Ydc is the lumbar intervertebral disc roundness deviation in the image. Repeat the process of obtaining medical image similarity between the sample analysis image and the sample historical image, obtain the medical image similarity between the sample analysis image and each historical analysis image, obtain multiple medical image similarities, and set the medical image similarity with the largest value as the image diagnosis rationality index corresponding to the sample analysis image. Repeat the process of obtaining the reasonable index of image diagnosis corresponding to the sample analysis image, and obtain the reasonable index of image diagnosis corresponding to each medical analysis image respectively; Obtain the preset range of reasonable index for image diagnosis. If the reasonable index for image diagnosis is within the preset range, the corresponding medical analysis image is screened as a qualified medical image. If the reasonable index for image diagnosis is not within the preset range, the corresponding medical analysis image is screened as a medical image to be treated, thus obtaining the initial screening data for medical images. It should be noted here that: The system acquires historical qualified medical images that have been screened by AI analysis tools, obtains the image diagnosis rationality index corresponding to each historical qualified medical image, compares the values of multiple image diagnosis rationality indices, sets the image diagnosis rationality index with the largest value as the first feature image rationality index, sets the image diagnosis rationality index with the smallest value as the second feature image rationality index, and sets the value range composed of the first feature image rationality index and the second feature image rationality index as the preset range of image diagnosis rationality index. In this application, qualified medical images include those where the reasonable index for image diagnosis is within the boundary of a preset range for the reasonable index for image diagnosis.
[0031] It should be noted here that: Step S2 above assesses the rationality of image diagnoses in medical image data by combining historical image diagnostic big data from medical institutions. Based on the assessment results, AI models are used to manage medical images with unreasonable diagnoses. The medical images with reasonable diagnoses output by the AI model are used to train the medical diagnostic model. A multi-dimensional assessment system is established based on massive historical image data, which can accurately identify abnormal image samples with anatomical structural abnormalities, contradictory diagnostic conclusions, or deviations in technical parameters, thus consolidating the foundation of training quality from the data source. A dynamic governance mechanism is constructed to achieve closed-loop optimization of "evaluation-governance-iteration", which not only effectively eliminates the interference of low-quality image data on model training, but also improves the accuracy of AI governance strategies through continuous feedback optimization. The high-quality diagnostic medical images ultimately adopted have stronger representativeness of pathological features and consistency of diagnostic logic. They can significantly enhance the medical diagnostic model's ability to deeply analyze image features and its generalization and adaptation capabilities across devices and diseases, ultimately achieving a systematic improvement in the accuracy and clinical usability of image diagnosis.
[0032] Step S3: Perform medical data processing based on the initial screening data of medical images and the data collected by medical institutions to obtain qualified data; Step S3 further includes the following steps: Acquire data collected by medical institutions, and obtain preliminary screening data for medical texts based on the data collected by medical institutions; Multiple medical texts to be processed are obtained based on the initial screening data of medical texts. These texts are then fed back to the text upload terminal, where medical staff correct them to obtain the texts to be judged. A text processing judgment model is then created. If the text processing judgment model screens the texts to be judged as qualified medical texts, the processing of the texts to be processed is complete. If the text processing judgment model screens the texts to be judged as unprocessable medical texts, the processing of the texts to be judged is repeated until the text processing judgment model screens the texts to be judged as qualified medical texts. The text governance judgment model is created as follows: Obtain multiple historical medical analysis texts, repeat step S1 to mark the historical medical analysis texts as medical texts to be treated and qualified medical texts, obtain medical text labeling data, and divide the medical text labeling data into medical text training set and medical text test set. It should be noted here that: In this application, the image training-to-test ratio is specifically set to 7:3, that is, the ratio of the number of medical cable infrared images in the medical text training set to the number of medical text test sets is 7:3. A text recognition model is created using an existing artificial intelligence platform. The model is then trained using a medical text training set until each medical cable infrared image in the medical text training set is used to train the text recognition model once. The text recognition model is tested using a medical text test set, and the recognition accuracy is obtained. When the recognition accuracy is greater than or equal to the target recognition accuracy, the text recognition model training is complete, and a text governance judgment model is obtained. When the recognition accuracy is less than the target recognition accuracy, the text recognition model is trained again using the medical text training set until the recognition accuracy is greater than or equal to the target recognition accuracy.
[0033] It should be noted here that: In this application, the target identification accuracy is 95%.
[0034] Multiple medical images to be treated are obtained based on the initial screening data of medical images. These images are then fed back to the image upload terminal, where medical staff correct them to obtain images to be judged. An image treatment judgment model is then created. If the image treatment judgment model screens the images to be judged as qualified images, the treatment of the images to be treated is completed. If the image treatment judgment model screens the images to be judged as images to be treated, the treatment of the images to be treated is repeated until the image treatment judgment model screens the images to be judged as qualified images. The image governance judgment model is created as follows: Acquire multiple historical medical analysis images, repeat step S2 to mark the historical medical analysis images as medical images to be treated and qualified medical images to obtain medical image labeling data, and divide the medical image labeling data into medical image training set and medical image test set. It should be noted here that: In this application, the image training-to-test ratio is specifically set to 7:3, that is, the ratio of the number of medical cable infrared images in the medical image training set to the number of medical image test sets is 7:3. An image recognition model is created using an existing artificial intelligence platform. The model is then trained using a medical image training set until each medical cable infrared image in the training set is used to train the image recognition model once. The image recognition model is tested using a medical image test set, and the recognition accuracy is obtained. When the recognition accuracy is greater than or equal to the target recognition accuracy, the training of the image recognition model is completed, and the image governance judgment model is obtained. When the recognition accuracy is less than the target recognition accuracy, the image recognition model is trained again using the medical image training set until the recognition accuracy is greater than or equal to the target recognition accuracy.
[0035] It should be noted here that: In this application, the target identification accuracy is 95%.
[0036] Define qualified medical images and qualified medical texts as qualified data for governance; It should be noted here that: In this application, the governance-compliant data referred to herein is used for model training of an AI analysis tool, which is a medical diagnostic model.
[0037] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.
Claims
1. An AI data governance method for medical institutions, characterized in that, Includes the following steps: Step S1: Acquire medical text data and medical image data, divide the medical text data into independent medical analysis texts, perform diagnostic rationality analysis on the medical analysis texts, screen the medical analysis texts according to the analysis results, obtain the initial screening data of medical texts, and define the initial screening data of medical texts and medical image data as data collected by medical institutions. Step S2: Perform diagnostic matching analysis on medical analysis images based on data collected by medical institutions, obtain the image diagnostic rationality index corresponding to the medical analysis images based on the analysis results, and screen the medical analysis images based on the image diagnostic rationality index to obtain the initial screening data of medical images. Step S3: Perform medical data processing based on the initial screening data of medical images and the data collected by medical institutions to obtain qualified data.
2. The AI data governance method for medical institutions according to claim 1, characterized in that, Step S1 further includes the following steps: Step S11: Obtain medical institutions for medical data governance, and select a target medical institution from the obtained medical institutions; Step S12: Obtain the medical text generated by the target medical institution at the current moment to obtain medical text data; obtain the medical images generated by the target medical institution at the current moment to obtain medical image data. Step S13: Conduct a governance needs analysis on the medical text data, and obtain the initial screening data of medical texts based on the analysis results; Step S14: Define the initial screening data of medical texts and medical image data as data collected by medical institutions.
3. The AI data governance method for medical institutions according to claim 2, characterized in that, Step S13 further includes the following steps: Step S131: Divide the medical text data into multiple medical analysis texts, and arbitrarily select one sample analysis text from the multiple medical analysis texts; Step S132: Perform symptom diagnosis matching analysis on the sample analysis text to obtain the symptom diagnosis matching degree corresponding to the sample analysis text; Step S133: Perform indicator diagnostic matching analysis on the sample analysis text to obtain the indicator diagnostic matching degree corresponding to the sample analysis text; Step S134: Calculate the text diagnosis rationality index corresponding to the sample analysis text by measuring the symptom diagnosis matching degree and the indicator diagnosis matching degree; Step S135: Obtain the text diagnosis rationality index corresponding to each medical analysis text; Step S136: Obtain the preset range of the reasonable index for text diagnosis. If the reasonable index for text diagnosis is within the preset range, the corresponding medical analysis text is screened as qualified medical text. If the reasonable index for text diagnosis is not within the preset range, the corresponding medical analysis text is screened as medical text to be treated, thus obtaining the initial screening data of medical text.
4. The AI data governance method for medical institutions according to claim 3, characterized in that, Step S132 further includes the following steps: The clinical diagnosis is obtained from the sample analysis text to obtain the target clinical diagnosis; The clinical symptoms of patients involved in the sample analysis text were obtained, resulting in the clinical symptoms of multiple patients; Multiple typical diagnostic symptoms were obtained, and the same symptoms in the patient's clinical symptoms and typical diagnostic symptoms were obtained to obtain the Z1 common clinical symptoms to the Za common clinical symptoms. The number of patients with the target clinical diagnosis who were treated at the target medical institution was statistically analyzed to obtain the cumulative number of historical target diagnoses. The number of historical target patients with common clinical symptoms of Z1 is statistically analyzed to obtain the number of Z1 symptom cases. The ratio of the number of Z1 symptom cases to the cumulative number of historical target diagnoses is calculated to obtain the historical occurrence percentage of Z1 symptoms. Similarly, the historical occurrence percentage of Za symptoms is calculated. Obtain the number of symptom cases corresponding to each typical symptom diagnosis, calculate the ratio of the obtained number of symptom cases to the cumulative number of historical target diagnoses, obtain the historical occurrence percentage corresponding to each typical symptom, and sum the obtained multiple historical occurrence percentages to obtain the cumulative occurrence percentage of typical symptoms. The symptom diagnosis matching degree corresponding to the sample analysis text is calculated by taking the percentage of historical occurrences of Z1 symptoms, the percentage of historical occurrences of Za symptoms, and the cumulative percentage of occurrences of typical symptoms.
5. The AI data governance method for medical institutions according to claim 3, characterized in that, Step S133 further includes the following steps: The clinical diagnosis is obtained from the sample analysis text to obtain the target clinical diagnosis; Abnormal medical indicators contained in the sample analysis text were obtained, resulting in multiple abnormal indicators for patients. The typical medical indicators involved in the target clinical diagnosis are obtained, resulting in multiple typical diagnostic indicators. The abnormal indicators of patients and the common medical indicators in the typical diagnostic indicators are obtained, resulting in Z1 common medical indicators to Zb common medical indicators. Acquire multiple historical target patients and perform statistical analysis to obtain the cumulative number of historical target diagnoses. Perform statistical analysis on historical target patients with the Z1 common medical indicator to obtain the number of Z1 indicator cases. Calculate the ratio of the number of Z1 indicator cases to the cumulative number of historical target diagnoses to obtain the historical occurrence percentage of the Z1 indicator. Similarly, count the number of Zb indicator cases and calculate the historical occurrence percentage of the Zb indicator. Obtain the number of cases corresponding to each typical diagnostic indicator, and calculate the ratio of the obtained number of cases to the cumulative number of historical target diagnoses to obtain the historical occurrence percentage of each typical diagnostic indicator. Then, sum the obtained historical occurrence percentages to obtain the cumulative occurrence percentage of the typical indicator. The historical occurrence percentage of the Z1 indicator, the historical occurrence percentage of the Zb indicator, and the cumulative occurrence percentage of typical indicators are used to calculate the indicator diagnostic matching degree corresponding to the sample analysis text.
6. The AI data governance method for medical institutions according to claim 1, characterized in that, Step S2 further includes the following steps: Step S21: Obtain data collected by medical institutions, obtain medical image data based on the data collected by medical institutions, split the medical image data into several medical analysis images, and arbitrarily select one sample analysis image from the multiple medical analysis images obtained; Step S22: Perform diagnostic matching degree analysis on the sample analysis images to obtain the image diagnosis rationality index corresponding to the sample analysis images; Step S23: Obtain the reasonable index for image diagnosis and the preset range of the reasonable index for each medical analysis image. If the reasonable index for image diagnosis is within the preset range, the corresponding medical analysis image is screened as a qualified medical image. If the reasonable index for image diagnosis is not within the preset range, the corresponding medical analysis image is screened as a medical image to be treated, thus obtaining the initial screening data for medical images.
7. The AI data governance method for medical institutions according to claim 6, characterized in that, Step S22 further includes the following steps: Step S221: Obtain the medical diagnosis results corresponding to the sample analysis images to obtain the sample image diagnosis; Step S222: Acquire historical medical images for which the medical diagnosis result is a sample image diagnosis, obtain multiple historical analysis images, and arbitrarily select one sample historical image from the multiple acquired historical analysis images; Step S223: If the sample image is diagnosed as lumbar disc herniation, perform a similarity analysis between the sample analysis image and the sample historical image, and obtain the medical image similarity between the sample analysis image and the sample historical image based on the analysis results. Step S223 further includes the following steps: Adjust the sample analysis image and the sample historical image to the same image scaling ratio, and use the geometric center corresponding to the sample analysis image as the origin to create a plane rectangular coordinate system, thus obtaining the image plane rectangular coordinate system; Mark the vertebral body connections in the sample analysis images, and mark the historical vertebral body connections in the sample historical images.
8. The AI data governance method for medical institutions according to claim 7, characterized in that, In step S223, also Includes the following steps: In the Cartesian coordinate system of the image plane, the historical image of the sample is covered by the image of the sample analysis. If the length of the line connecting the vertebrae in the analysis is greater than or equal to that of the historical line connecting the vertebrae, the line connecting the vertebrae in the analysis is completely covered by the line connecting the historical line connecting the vertebrae. If the length of the line connecting the vertebrae in the sample is less than that of the historical line connecting the vertebrae, the line connecting the historical line connecting the vertebrae in the analysis is completely covered by the line connecting the vertebrae in the analysis. Pixel count ratio analysis was performed on the lumbar intervertebral disc region in the sample analysis images and the sample historical images, and the roundness of the first intervertebral disc and the second intervertebral disc were obtained based on the analysis results.
9. The AI data governance method for medical institutions according to claim 8, characterized in that, Step S223 further includes the following steps: Calculate the difference between the roundness of the first intervertebral disc and the roundness of the second intervertebral disc, and calculate the ratio of the obtained difference to the roundness of the second intervertebral disc to obtain the lumbar disc roundness deviation in the imaging. The overlapping area between the first and second lumbar intervertebral disc regions is obtained to obtain the third lumbar intervertebral disc region. The number of pixels in the first, second, and third lumbar intervertebral disc regions is counted to obtain the pixel count values of the first region, the second region, and the third region. The overlap rate of the intervertebral disc region in the image is obtained by calculating the pixel count values of the first region, the second region, and the third region. The medical image similarity between the sample analysis image and the sample historical image is obtained by calculating the overlap rate of the lumbar intervertebral disc region and the lumbar intervertebral disc roundness deviation. Obtain the medical image similarity between the sample analysis image and each historical analysis image, obtain multiple medical image similarities, and set the medical image similarity with the largest value as the image diagnosis rationality index corresponding to the sample analysis image.
10. The AI data governance method for medical institutions according to claim 1, characterized in that, Step S3 further includes the following steps: Acquire data collected by medical institutions, and obtain preliminary screening data for medical texts based on the data collected by medical institutions; Multiple medical texts to be processed are obtained based on the initial screening data of medical texts. These texts are then fed back to the text upload terminal for correction, resulting in medical texts to be judged. A text processing judgment model is then created. If the text processing judgment model screens the medical texts to be judged as qualified medical texts, the processing of the medical texts to be processed is complete. If the text processing judgment model screens the medical texts to be judged as medical texts to be processed, the processing of the medical texts to be processed is repeated until the text processing judgment model screens the medical texts to be judged as qualified medical texts. Multiple medical images to be treated are obtained based on the initial screening data of medical images, and the medical images to be treated are fed back to the image upload terminal for correction, resulting in medical images to be judged. An image treatment judgment model is created. If the image treatment judgment model screens the medical images to be judged into qualified medical images, the treatment of the medical images to be treated is completed. If the image treatment judgment model screens the medical images to be judged into medical images to be treated, the treatment of the medical images to be treated is repeated until the image treatment judgment model screens the medical images to be judged into qualified medical images. Qualified medical images and qualified medical texts are defined as qualified data for governance.