Intelligent auxiliary diagnosis decision system and control method based on AI chip

Abstract

The present invention discloses an intelligent auxiliary diagnosis decision system and control method based on an AI chip, which relates to the field of intelligent auxiliary diagnosis technology, including an intelligent handheld auxiliary diagnosis terminal based on an AI chip, a terminal wireless transmission module, and a server-side medical large model analysis module. The present invention can use an intelligent handheld auxiliary diagnosis terminal based on an AI chip to collect health-related data of patients, an edge-side auxiliary diagnosis analysis module based on an AI chip to parse the original data describing the health status information, and perform edge-side health status diagnosis and analysis based on the original data describing the health status information; the data source of the server-side medical large model analysis module can also be combined with historical medical data, and the patient's terminal comprehensive health data can be combined to output the patient's health assessment results, realizing the initial diagnosis from the edge, patient self-screening, to the intelligent auxiliary diagnosis on the server side, and finally providing the attending physician with more accurate intelligent auxiliary diagnosis options.

Description

Technical Field

[0001] The present invention relates to the field of intelligent auxiliary diagnosis technology, and in particular to an intelligent auxiliary diagnosis decision system and a control method based on an AI chip. Background Art

[0002] The application of artificial intelligence (AI) technology in medical diagnosis, especially deep learning and machine learning algorithms, has made significant progress in analyzing medical images and assisting doctors in making more accurate diagnoses. For example, Google DeepMind's AI system has reached expert level in analyzing ophthalmic diseases.

[0003] Clinical decision support: The application of AI in clinical decision support systems (CDSS) provides doctors with personalized treatment recommendations by integrating patients’ medical records, laboratory results, and the latest medical research, thereby improving the efficiency and accuracy of diagnosis and treatment.

[0004] Medical image-assisted diagnosis: AI medical image-assisted diagnosis systems have developed rapidly both at home and abroad, including in terms of financing, policy support, demand and industry competition landscape.

[0005] Policy support: Since 2015, a number of policies have been introduced to promote the development of medical AI, which have played a guiding role in the application of AI in the medical field.

[0006] Artificial intelligence is also widely used in the field of patient health tracking and auxiliary diagnosis. It is usually implemented by installing relevant smart tracking apps on smartphones. It can usually collect preliminary health information of patients, and then make preliminary health diagnoses based on the setting of some thresholds. The technical route of this auxiliary diagnosis lacks accuracy and reliability, so the effect is not ideal and it is difficult to be recognized by patient users.

[0007] In summary, intelligent assisted diagnosis technology has significant application potential in the medical field, but it also faces many challenges and defects that need to be solved through further technological innovation and policy support. Summary of the invention

[0008] In order to solve the above-mentioned technical problems of intelligent assisted diagnosis, the present invention provides an intelligent assisted diagnosis decision system and control method based on AI chip. The following technical solutions are adopted:

[0009] An AI chip-based intelligent auxiliary diagnosis decision system includes an AI chip-based intelligent handheld auxiliary diagnosis terminal, a terminal wireless transmission module and a server-side medical large model analysis module. The intelligent handheld auxiliary diagnosis terminal includes an intelligent terminal and an edge-side auxiliary diagnosis analysis module based on an AI chip. The intelligent terminal receives the patient's health status information description, the edge-side auxiliary diagnosis analysis module is connected to the intelligent terminal for communication, collects the original data of the health status information description, parses the original data of the health status information description, performs edge-side health status diagnosis and analysis based on the original data of the health status information description, interacts with the intelligent terminal to obtain the edge-side health status diagnosis and analysis results, and displays the edge-side health status diagnosis and analysis results through the intelligent terminal. The patient eliminates several diagnostic results that do not meet the requirements of his or her own health status by operating the intelligent terminal according to the edge-side health status diagnosis and analysis results. The intelligent terminal packages the remaining diagnostic results, timestamps and patient information tag data to form a comprehensive health data packet for the patient terminal, and transmits the comprehensive health data packet for the patient terminal to the server-side medical large model analysis module through the terminal wireless transmission module;

[0010] The server-side medical big model analysis module is equipped with a health diagnosis big model based on a machine learning architecture. The health assistance big model combines historical medical data and the patient's terminal comprehensive health data to output the patient's health assessment results.

[0011] Optionally, the intelligent terminal includes a camera module, a microphone module, a memory, a visual analysis chip, an audio analysis chip and a processor. The camera module is used to shoot the patient's health examination video data, and the microphone module is used to collect the patient's self-reported health condition voice data. The camera module, microphone module, visual analysis chip and audio analysis chip are respectively communicated with the memory. The microphone module pre-stores health condition data to collect guide video data. The visual analysis chip uses text recognition technology and keyword extraction technology to obtain patient health examination data from health examination video data and stores it in the memory. The microphone module uses voice recognition technology to analyze the health condition self-reported voice data to obtain patient health-related data and stores it in the memory. The processor is communicated with the memory to fuse the patient's health examination data and the patient's health-related data to generate health status information describing original data.

[0012] Optionally, the intelligent terminal further includes a touch display screen, and the processor controls the touch display screen to display health status data collection guidance video data, and plays it in segments according to the patient's operation steps.

[0013] Optionally, the edge-assisted diagnosis and analysis module includes an edge data storage device and an AI chip. The edge data storage device communicates and exchanges health status information description raw data with the storage device. The AI ​​chip is communicatively connected to the edge data storage device, performs edge health status diagnosis and analysis on the raw data describing the health status information, and stores the analysis results in the edge data storage device. The edge data storage device and the storage exchange edge health status diagnosis and analysis results, and display the edge health status diagnosis and analysis results through a touch screen. The patient eliminates several diagnostic results that are inconsistent with his or her own health status by operating the touch screen. The processor packages the remaining diagnostic results and adds timestamps and patient information tag data to form a comprehensive health data packet for the patient terminal. The patient's comprehensive health data packet is transmitted to the server-side medical large model analysis module through the terminal wireless transmission module.

[0014] Optionally, the AI ​​chip is a processor with an integrated neural network processing unit.

[0015] By adopting the above technical solution, an AI chip-based intelligent handheld auxiliary diagnosis terminal is used to collect the patient's health-related data. The specific intelligent terminal can collect health examination video data through the camera module. The health examination video data refers to the video data of the patient scanning different health examination forms under the camera module. Usually, the health examination form includes a blood test report, a urine test report, a stool test report, an electrocardiogram report, and an imaging test report, etc. These checklists can provide accurate data for intelligent auxiliary diagnosis. The intelligent terminal can also collect the patient's health self-report voice data through the microphone module. The health self-report voice data is collected in the form of questions and answers. A simple and intuitive user interface is designed, and a health self-report guidance video in the form of questions and answers is designed to guide the patient's answer to health-related data. For example, the opening remarks introduce the purpose and process of the question and answer to ensure that the patient is informed and agrees. Questions are presented to the patient one by one, which can be in the form of text, voice or video. The patient answers the questions by voice. After the patient completes the answer, the completeness and rationality of the answer can be automatically checked. If it is incomplete, the question and answer can be supplemented.

[0016] The processor fuses the patient's health examination data and the patient's health-related data to generate raw data describing the health status information, which provides raw data for subsequent AI intelligent diagnosis.

[0017] The edge-assisted diagnosis and analysis module based on the AI ​​chip parses the original data describing the health status information, and performs edge-assisted health status diagnosis and analysis based on the original data describing the health status information, realizing preliminary intelligent auxiliary diagnosis on the edge. Here, a processor with an integrated neural network processing unit is used to realize neural network processing, which is highly efficient and can give preliminary diagnosis results and corresponding probabilities. Generally, no more than ten diagnosis results are provided, which are displayed to patients through a touch screen. Since it is a preliminary diagnosis on the edge without the support of massive medical diagnosis data on the server, there will be some diagnostic results with large discrepancies. For example, a patient has a headache caused by high blood pressure, but the diagnosis result shows a headache caused by a cold. After seeing the corresponding edge diagnosis result, the patient does not have obvious cold symptoms according to his own situation, so he can choose to actively remove the diagnosis result of the cold, which can further reduce the amount of auxiliary diagnosis calculations on the subsequent server and improve the accuracy of intelligent auxiliary diagnosis decisions;

[0018] The processor packages the remaining diagnosis results, timestamps and patient information tag data into a comprehensive health data packet of the patient terminal, and transmits the comprehensive health data packet of the patient terminal to the server-side medical big model analysis module through the terminal wireless transmission module; the data source of the server-side medical big model analysis module can also be combined with historical medical data, combined with the comprehensive health data of the patient terminal to output the patient's health assessment results, and use historical medical data to train a medical big model, such as a deep learning network. The model can be a convolutional neural network (CNN), a recurrent neural network (RNN), or a Transformer, etc. Use the trained medical big model to evaluate the patient's comprehensive health data.

[0019] Assuming that the historical medical data is (H) and the patient terminal health data is (P), the integrated data set (D) is: [D = H + P];

[0020] Each sample in the dataset is represented as a feature vector s. Assume that the model (f) is a deep learning network and use the Softmax function to output the probabilities of different diagnosis results.

[0021] To build a server-side medical big model analysis module to output the patient's health assessment results.

[0022] It realizes preliminary diagnosis on the edge, patient self-screening, to intelligent assisted diagnosis on the server, and ultimately provides more accurate intelligent assisted diagnosis options for attending physicians.

[0023] Optionally, the terminal wireless transmission module is a 4G / 5G wireless transmission module.

[0024] Optionally, the server-side medical big model analysis module includes a server-side wireless transmission module and a computer, wherein the computer receives a patient terminal comprehensive health data packet sent by the terminal wireless transmission module through the server-side wireless transmission module, and enters the patient terminal comprehensive health data packet into a health diagnosis big model based on a machine learning architecture after parsing the patient terminal comprehensive health data packet. The health diagnosis big model combines historical medical data and the patient terminal comprehensive health data to output the patient's health assessment results. The patient's health assessment results are used to assist doctors in health diagnosis. The doctor inputs the diagnosis results into the computer, and the diagnosis results are transmitted to the smart terminal for display through the server-side wireless transmission module and the terminal wireless transmission module.

[0025] By adopting the above technical solutions, doctors can combine the auxiliary decisions given by the intelligent auxiliary diagnosis decision-making system and rely on their own experience to finally determine the health status and cause of the patient, thereby achieving more accurate AI plus experience intelligent diagnosis results and providing an accurate data basis for the formulation of subsequent treatment plans or rehabilitation plans.

[0026] The intelligent auxiliary diagnosis decision control method based on AI chip adopts the intelligent terminal of the intelligent auxiliary diagnosis decision system based on AI chip for control, and includes the following steps:

[0027] Step 1: The patient operates the touch screen to select an auxiliary diagnosis option;

[0028] Step 2, the processor retrieves health data collection guidance video data, the health data collection guidance video data includes a welcome interface video segment, a test data entry guidance video segment, and a question-and-answer form health self-report guidance video segment;

[0029] Step 3, the processor controls the touch display screen to first play the welcome interface video segment, and then play the examination data entry guidance video segment. The examination data entry guidance video segment provides examination data entry options, and the examination data entry options include shooting entry options, local file entry options, and examination data entry completion options. When the patient selects the shooting entry option, the camera module is called to shoot. When the patient selects the local file entry option, the memory is scanned to provide the patient with picture or video selection;

[0030] Step 4, when the patient selects the option of completing the entry of examination data, the processor controls the touch screen to play a video segment in the form of a question and answer to guide the self-reporting of health conditions. After each question in the video segment in the form of a question and answer to guide the self-reporting of health conditions is completed, the microphone module is called to collect the patient's self-reporting voice data;

[0031] Step 5: When the self-reporting voice data of the patient in the last item of the question-and-answer form health condition self-reporting guidance video segment is collected, the original data collection of the health status information description of the smart terminal is completed;

[0032] Step 6: The edge data storage device communicates with the storage device to exchange the original data of the health status information description. The AI ​​chip performs edge health status diagnosis and analysis on the original data of the health status information description based on the deep learning model. The analysis results of the edge health status diagnosis and analysis include multiple diagnosis results and the probabilities corresponding to the diagnosis results. The edge health status diagnosis and analysis results are displayed on the touch screen. The patient, based on his or her basic understanding of his or her own situation, eliminates several diagnosis results that are determined to be inconsistent with his or her own health status by operating the touch screen.

[0033] Step 7, the processor packages the remaining diagnostic results and adds the timestamp and patient information tag data to form a comprehensive health data packet for the patient terminal, and transmits the comprehensive health data packet for the patient terminal to the server-side medical large model analysis module through the terminal wireless transmission module.

[0034] By adopting the above-mentioned technical solution and using segmented controlled playback of health data collection guidance videos, it is possible to guide patients to complete the efficient and complete collection of examination form information. It is also possible to guide patients to actively speak out key health-related content or data in a question-and-answer format, thereby providing accurate original data for subsequent edge intelligent analysis and diagnosis.

[0035] Optionally, in step 6, the AI ​​chip performs edge health status diagnosis and analysis on the raw data describing the health status information based on the deep learning model, including the following sub-steps:

[0036] Step 61, digitizing the patient health examination data and the patient health-related data;

[0037] Step 62, normalizing or standardizing the data;

[0038] Step 63, merging the feature vectors of the health condition question and answer data corresponding to different types of examination list data and patient health-related data into a comprehensive feature vector X;

[0039] In step 64, the comprehensive feature vector X is input into the neural network of the AI ​​chip. The output layer of the neural network of the AI ​​chip uses a softmax activation function to output multiple diagnosis results and corresponding probabilities, and the top ten diagnosis results and corresponding probabilities are selected for display.

[0040] By adopting the above technical solution, the neural network of the AI ​​chip can efficiently realize the diagnosis and analysis of the patient's health status based on the softmax activation function, and output the probability corresponding to each diagnostic result respectively.

[0041] Optionally, for each health status information describing raw data i and each diagnosis result category j, the softmax function is calculated as follows:

[0042]

[0043] Where P(y i |X i ) indicates that in the feature vector X i The probability of diagnosis result category j under the input, e is the base of natural logarithm, Z ij is the eigenvector X i The raw score corresponding to the diagnosis category j, Denotes the feature vector X i Corresponding to the raw score for diagnosis result category k in the neural network, m is the total number of diagnosis result categories.

[0044] In summary, the present invention includes at least one of the following beneficial technical effects:

[0045] The present invention can provide an intelligent auxiliary diagnosis decision system and control method based on an AI chip, adopt an intelligent handheld auxiliary diagnosis terminal based on an AI chip to collect health-related data of patients, and the processor fuses the patient's health examination data and the patient's health-related data to generate health status information description raw data to provide raw data for subsequent AI intelligent diagnosis; the edge-side auxiliary diagnosis analysis module based on the AI ​​chip parses the health status information description raw data, and performs edge-side health status diagnosis analysis based on the health status information description raw data, thereby realizing preliminary intelligent auxiliary diagnosis at the edge, and being able to give preliminary diagnosis results and corresponding probabilities, which are displayed to patients through a touch display screen. After seeing the corresponding edge-side diagnosis results, the patients actively remove the diagnosis results that do not match themselves, further reducing the auxiliary diagnosis calculation amount of the subsequent server side, and improving the accuracy of intelligent auxiliary diagnosis decisions;

[0046] The processor packages the remaining diagnostic results, timestamps and patient information tag data into a comprehensive health data packet for the patient terminal, and transmits the comprehensive health data packet for the patient terminal to the server-side medical big model analysis module through the terminal wireless transmission module; the data source of the server-side medical big model analysis module can also be combined with historical medical data and the patient terminal comprehensive health data to output the patient's health assessment results, realizing preliminary diagnosis at the edge, patient self-screening, to intelligent assisted diagnosis on the server side, and ultimately providing the attending physician with more accurate intelligent assisted diagnosis options. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a schematic diagram of the component connection principle of the intelligent auxiliary diagnosis and decision-making system based on the AI ​​chip of the present invention;

[0048] Figure 2 It is a schematic diagram of the connection principle of the intelligent handheld auxiliary diagnosis terminal components based on the AI ​​chip intelligent auxiliary diagnosis decision system of the present invention;

[0049] Figure 3 It is a flow chart of the intelligent auxiliary diagnosis decision control method based on AI chip of the present invention.

[0050] Explanation of the accompanying drawings: 1. Intelligent handheld auxiliary diagnosis terminal; 11. Intelligent terminal; 111. Camera module; 112. Microphone module; 113. Memory; 114. Visual analysis chip; 115. Audio analysis chip; 116. Processor; 12. Edge-assisted diagnosis and analysis module; 121. Edge data storage device; 122. AI chip; 2. Terminal wireless transmission module; 3. Server-side medical large model analysis module; 31. Server-side wireless transmission module; 32. Computer. DETAILED DESCRIPTION

[0051] The present invention is further described in detail below in conjunction with the accompanying drawings.

[0052] The embodiments of the present invention disclose an intelligent auxiliary diagnosis decision system and a control method based on an AI chip.

[0053] Reference Figure 1-Figure 3 , Embodiment 1, an intelligent auxiliary diagnosis decision system based on an AI chip, comprising an intelligent handheld auxiliary diagnosis terminal 1 based on an AI chip, a terminal wireless transmission module 2 and a server-side medical large model analysis module 3, the intelligent handheld auxiliary diagnosis terminal 1 comprises an intelligent terminal 11 and an edge-side auxiliary diagnosis analysis module 12 based on an AI chip, the intelligent terminal 11 receives the patient's health status information description, the edge-side auxiliary diagnosis analysis module 12 is connected to the intelligent terminal 11 in communication, collects the original data of the health status information description, parses the original data of the health status information description, performs edge-side health status diagnosis and analysis based on the original data of the health status information description, interacts with the intelligent terminal 11 to obtain the edge-side health status diagnosis and analysis results, and displays the edge-side health status diagnosis and analysis results through the intelligent terminal 11, the patient eliminates several diagnosis results that do not meet the requirements of his own health status by operating the intelligent terminal 11 according to the edge-side health status diagnosis and analysis results, the intelligent terminal 11 packages the remaining diagnosis results, timestamps and patient information tag data into a comprehensive health data packet of the patient terminal, and transmits the comprehensive health data packet of the patient terminal to the server-side medical large model analysis module 3 through the terminal wireless transmission module 2;

[0054] The server-side medical big model analysis module 3 is equipped with a health diagnosis big model based on a machine learning architecture. The health assistance big model combines historical medical data and the patient's terminal comprehensive health data to output the patient's health assessment results.

[0055] Embodiment 2, the intelligent terminal 11 includes a camera module 111, a microphone module 112, a memory 113, a visual analysis chip 114, an audio analysis chip 115 and a processor 116, the camera module 111 is used to shoot the patient's health examination video data, the microphone module 112 is used to collect the patient's health condition self-report voice data, the camera module 111, the microphone module 112, the visual analysis chip 114 and the audio analysis chip 115 are respectively communicated with the memory 113, the microphone module 112 pre-stores the health condition data collection guide video data, the visual analysis chip 114 uses text recognition technology and keyword extraction technology to obtain the patient's health examination data from the health examination video data and stores it in the memory 113, the microphone module 112 uses voice recognition technology to analyze the health condition self-report voice data to obtain the patient's health-related data and store it in the memory 113, the processor 116 is communicated with the memory 113, and the patient's health examination data and the patient's health-related data are merged to generate the original data describing the health status information.

[0056] In Example 3, the intelligent terminal 11 further includes a touch screen 117, and the processor 116 controls the touch screen 117 to display health status data collection guidance video data, and plays it in segments according to the patient's operation steps.

[0057] Embodiment 4, the edge auxiliary diagnosis and analysis module 12 includes an edge data storage device 121 and an AI chip 122. The edge data storage device 121 communicates and exchanges with the memory 113 to describe the original data of the health status information. The AI ​​chip 122 is communicatively connected to the edge data storage device 121, performs edge health status diagnosis and analysis on the original data describing the health status information, and stores the analysis results in the edge data storage device 121. The edge data storage device 121 exchanges the edge health status diagnosis and analysis results with the memory 113, and displays the edge health status diagnosis and analysis results through the touch screen 117. The patient eliminates several diagnosis results that are determined to be inconsistent with his or her own health status by operating the touch screen 117. The processor 116 packages the remaining diagnosis results and adds a timestamp and patient information tag data to form a comprehensive health data packet of the patient terminal, and transmits the comprehensive health data packet of the patient terminal to the server-side medical large model analysis module 3 through the terminal wireless transmission module 2.

[0058] In Embodiment 5, the AI ​​chip 122 is a processor that integrates a neural network processing unit.

[0059] The AI ​​chip-based intelligent handheld auxiliary diagnosis terminal 1 is used to collect the patient's health-related data. The specific intelligent terminal 11 can collect health examination video data through the camera module 111. The health examination video data refers to the video data of the patient scanning different health examination forms under the camera module 111. Usually, the health examination form includes a blood test report, a urine test report, a stool test report, an electrocardiogram report, and an imaging test report, etc. These checklists can provide accurate data for intelligent auxiliary diagnosis. The intelligent terminal 11 can also collect the patient's health self-report voice data through the microphone module 112. The health self-report voice data is collected in the form of questions and answers. A simple and intuitive user interface is designed, and a health self-report guidance video in the form of questions and answers is designed to guide the patient's answer to health-related data. For example, the opening remarks introduce the purpose and process of the question and answer to ensure that the patient is informed and agrees. Questions are presented to the patient one by one, which can be in the form of text, voice or video. The patient answers the questions by voice. After the patient completes the answer, the completeness and rationality of the answer can be automatically checked. If it is incomplete, the question and answer can be continued.

[0060] The processor 116 fuses the patient health examination data and the patient health-related data to generate raw data describing health status information to provide raw data for subsequent AI intelligent diagnosis;

[0061] The edge-side auxiliary diagnosis and analysis module 12 based on the AI ​​chip parses the original data describing the health status information, and performs edge-side health status diagnosis and analysis based on the original data describing the health status information, thereby realizing preliminary intelligent auxiliary diagnosis at the edge. Here, a processor with an integrated neural network processing unit is used to realize neural network processing, which is highly efficient and can give preliminary diagnosis results and corresponding probabilities. Generally, no more than ten diagnosis results are provided, which are displayed to the patient through the touch display screen 117. Since it is a preliminary diagnosis at the edge without the support of massive medical diagnosis data on the server side, there will be some diagnostic results with large discrepancies. For example, a patient has a headache caused by high blood pressure, but the diagnosis result shows a headache caused by a cold. After seeing the corresponding edge diagnosis result, the patient does not have obvious cold symptoms according to his own situation, so he can choose to actively remove the diagnosis result of the cold, which can further reduce the auxiliary diagnosis calculation amount of the subsequent server side and improve the accuracy of intelligent auxiliary diagnosis decision-making;

[0062] The processor 116 packages the remaining diagnosis results, timestamps and patient information tag data into a comprehensive health data packet of the patient terminal, and transmits the comprehensive health data packet of the patient terminal to the server-side medical big model analysis module 3 through the terminal wireless transmission module 2; the data source of the server-side medical big model analysis module 3 can also be combined with historical medical data, combined with the comprehensive health data of the patient terminal to output the health assessment result of the patient, and use the historical medical data to train a medical big model, such as a deep learning network. The model can be a convolutional neural network (CNN), a recurrent neural network (RNN), or a Transformer, etc. Use the trained medical big model to evaluate the patient's comprehensive health data.

[0063] Assuming that the historical medical data is (H) and the patient terminal health data is (P), the integrated data set (D) is: [D = H + P];

[0064] Each sample in the dataset is represented as a feature vector s. Assume that the model (f) is a deep learning network and use the Softmax function to output the probabilities of different diagnosis results.

[0065] To build a server-side medical big model analysis module to output the patient's health assessment results.

[0066] It realizes preliminary diagnosis on the edge, patient self-screening, to intelligent assisted diagnosis on the server, and ultimately provides more accurate intelligent assisted diagnosis options for attending physicians.

[0067] In embodiment 6, the terminal wireless transmission module 2 is a 4G / 5G wireless transmission module.

[0068] Example 7, the server-side medical big model analysis module 3 includes a server-side wireless transmission module 31 and a computer 32. The computer 32 receives the patient terminal comprehensive health data packet sent by the terminal wireless transmission module 2 through the server-side wireless transmission module 31, and enters the health diagnosis big model based on the machine learning architecture after parsing the patient terminal comprehensive health data packet. The health diagnosis big model combines historical medical data and patient terminal comprehensive health data to output the patient's health assessment results. The patient's health assessment results are used to assist doctors in health diagnosis. The doctor inputs the diagnosis results on the computer 32, and the diagnosis results are transmitted to the smart terminal 11 through the server-side wireless transmission module 31 and the terminal wireless transmission module 2 for display.

[0069] Doctors combine the auxiliary decisions given by the intelligent auxiliary diagnosis decision-making system with their own experience to ultimately determine the health status and cause of the patient, achieving more accurate AI plus experience-based intelligent diagnosis results, and providing an accurate data basis for the formulation of subsequent treatment plans or rehabilitation plans.

[0070] Embodiment 8, the intelligent auxiliary diagnosis decision control method based on AI chip, adopts the intelligent terminal 11 of the intelligent auxiliary diagnosis decision system based on AI chip for control, and comprises the following steps:

[0071] Step 1, the patient operates the touch display screen 117 to select an auxiliary diagnosis option;

[0072] Step 2, the processor 116 retrieves health data collection guidance video data, the health data collection guidance video data includes a welcome interface video segment, a test data entry guidance video segment, and a question-and-answer form health self-report guidance video segment;

[0073] Step 3, the processor 116 controls the touch screen 117 to first play the welcome interface video segment, and then play the examination data entry guidance video segment. The examination data entry guidance video segment provides examination data entry options, and the examination data entry options include a shooting entry option, a local file entry option, and an examination data entry completion option. When the patient selects the shooting entry option, the camera module 111 is called to shoot. When the patient selects the local file entry option, the memory 113 is scanned to provide the patient with a picture or video selection;

[0074] Step 4, when the patient selects the option of completing the entry of examination data, the processor 116 controls the touch screen 117 to play a video segment in the form of a question and answer to guide the self-reporting of health conditions. After each question in the video segment in the form of a question and answer to guide the self-reporting of health conditions is completed, the microphone module 112 is called to collect the patient's self-reporting voice data;

[0075] Step 5, when the self-reporting voice data of the patient in the last item of the question-and-answer form health condition self-reporting guidance video segment is collected, the original data collection of the health status information description of the smart terminal 11 is completed;

[0076] Step 6: The edge data storage 121 communicates with the storage 113 to exchange the original data describing the health status information. The AI ​​chip 122 performs edge health status diagnosis and analysis on the original data describing the health status information based on the deep learning model. The analysis results of the edge health status diagnosis and analysis include multiple diagnosis results and the probabilities corresponding to the diagnosis results. The edge health status diagnosis and analysis results are displayed through the touch screen 117. The patient, based on his or her basic understanding of his or her own situation, eliminates several diagnosis results that are determined to be inconsistent with his or her own health status by operating the touch screen 117.

[0077] In step 7, the processor 116 packages the remaining diagnostic results and adds the timestamp and patient information tag data to form a comprehensive health data packet for the patient terminal, and transmits the comprehensive health data packet for the patient terminal to the server-side medical large model analysis module 3 through the terminal wireless transmission module 2.

[0078] By adopting the segmented controlled playback of health data collection guidance videos, patients can be guided to complete the efficient and complete collection of examination form information. Patients can also be guided to actively speak out key health-related content or data in a question-and-answer format, providing accurate original data for subsequent edge intelligent analysis and diagnosis.

[0079] In Example 9, in step 6, the AI ​​chip 122 performs edge health status diagnosis and analysis on the raw data describing the health status information based on the deep learning model, including the following sub-steps:

[0080] Step 61, digitizing the patient health examination data and the patient health-related data;

[0081] Step 62, normalizing or standardizing the data;

[0082] Step 63, merging the feature vectors of the health condition question and answer data corresponding to different types of examination list data and patient health-related data into a comprehensive feature vector X;

[0083] In step 64, the comprehensive feature vector X is input into the neural network of the AI ​​chip 122. The output layer of the neural network of the AI ​​chip 122 uses a softmax activation function to output multiple diagnosis results and corresponding probabilities, and the top ten diagnosis results and corresponding probabilities are selected for display.

[0084] In Example 10, the neural network of the AI ​​chip 122 can efficiently perform diagnostic analysis on the patient's health status based on the softmax activation function, and output the probability corresponding to each diagnostic result.

[0085] For each health status information describing the original data i and each diagnosis result category j, the softmax function is calculated as follows:

[0086]

[0087] Where P(y i |X i ) indicates that in the feature vector X i The probability of diagnosis result category j under the input, e is the base of natural logarithm, Z ij is the eigenvector X i The raw score corresponding to the diagnosis category j, Denotes the feature vector X i Corresponding to the raw score for diagnosis result category k in the neural network, m is the total number of diagnosis result categories.

[0088] The above are all preferred embodiments of the present invention, and are not intended to limit the protection scope of the present invention. Therefore, any equivalent changes made based on the structure, shape, and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. Intelligent auxiliary diagnosis decision system based on AI chip, characterized by: The invention comprises an intelligent handheld auxiliary diagnosis terminal (1) based on an AI chip, a terminal wireless transmission module (2) and a server-side medical large model analysis module (3), wherein the intelligent handheld auxiliary diagnosis terminal (1) comprises an intelligent terminal (11) and an edge-side auxiliary diagnosis analysis module (12) based on an AI chip, wherein the intelligent terminal (11) receives a patient's health status information description, the edge-side auxiliary diagnosis analysis module (12) is connected in communication with the intelligent terminal (11), collects original data of the health status information description, parses the original data of the health status information description, performs edge-side health status diagnosis analysis based on the original data of the health status information description, exchanges edge-side health status diagnosis analysis results with the intelligent terminal (11), and displays the edge-side health status diagnosis analysis results through the intelligent terminal (11), wherein the patient eliminates a number of diagnosis results that do not conform to the patient's health status by operating the intelligent terminal (11) according to the edge-side health status diagnosis analysis results, the intelligent terminal (11) packages the remaining diagnosis results, timestamps and patient information tag data into a patient terminal comprehensive health data packet, and transmits the patient terminal comprehensive health data packet to the server-side medical large model analysis module (3) through the terminal wireless transmission module (2); The server-side medical big model analysis module (3) is equipped with a health diagnosis big model based on a machine learning architecture. The health assistance big model combines historical medical data and comprehensive health data of the patient terminal to output the patient's health assessment results.

2. The AI ​​chip-based intelligent auxiliary diagnosis decision system according to claim 1, characterized in that: The intelligent terminal (11) comprises a camera module (111), a microphone module (112), a memory (113), a visual analysis chip (114), an audio analysis chip (115) and a processor (116). The camera module (111) is used to shoot a patient's health examination video data. The microphone module (112) is used to collect the patient's self-reported health condition voice data. The camera module (111), the microphone module (112), the visual analysis chip (114) and the audio analysis chip (115) are respectively connected to the memory (113) in communication. The microphone module (112) pre-stores health condition data to collect guide video data. The visual analysis chip (114) uses text recognition technology and keyword extraction technology to obtain patient health examination data from the health examination video data and stores it in the memory (113). The microphone module (112) uses voice recognition technology to analyze the self-reported health condition voice data to obtain patient health-related data and stores it in the memory (113). The processor (116) is connected to the memory (113) in communication to fuse the patient's health examination data and the patient's health-related data to generate health status information description raw data.

3. The AI ​​chip-based intelligent auxiliary diagnosis decision system according to claim 2 is characterized in that: The intelligent terminal (11) also includes a touch display screen (117), and the processor (116) controls the touch display screen (117) to display health condition data collection guidance video data, and plays it in segments according to the patient's operation steps.

4. The AI ​​chip-based intelligent auxiliary diagnosis decision system according to claim 3 is characterized in that: The edge auxiliary diagnosis and analysis module (12) includes an edge data storage device (121) and an AI chip (122). The edge data storage device (121) communicates and exchanges with the memory (113) to exchange the original data of health status information description. The AI ​​chip (122) is connected to the edge data storage device (121) to perform edge health status diagnosis and analysis on the original data of health status information description, and stores the analysis result in the edge data storage device (121). The edge data storage device (121) and the memory (113) exchange the edge health status diagnosis and analysis result, and display the edge health status diagnosis and analysis result through the touch display screen (117). The patient eliminates a number of diagnosis results that are determined to be inconsistent with his or her own health status by operating the touch display screen (117). The processor (116) packages the remaining diagnosis results and adds a timestamp and patient information tag data to form a patient terminal comprehensive health data packet, and transmits the patient terminal comprehensive health data packet to the server-side medical large model analysis module (3) through the terminal wireless transmission module (2).

5. The AI ​​chip-based intelligent auxiliary diagnosis decision system according to claim 4 is characterized in that: The AI ​​chip (122) is a processor that integrates a neural network processing unit.

6. The AI ​​chip-based intelligent auxiliary diagnosis and decision-making system according to claim 5 is characterized in that: The terminal wireless transmission module (2) is a 4G / 5G wireless transmission module.

7. The AI ​​chip-based intelligent auxiliary diagnosis and decision-making system according to claim 6 is characterized in that: The server-side medical big model analysis module (3) comprises a server-side wireless transmission module (31) and a computer (32). The computer (32) receives the patient terminal comprehensive health data packet sent by the terminal wireless transmission module (2) through the server-side wireless transmission module (31), parses the patient terminal comprehensive health data packet and enters the data into a health diagnosis big model based on a machine learning architecture. The health diagnosis big model combines historical medical data and the patient terminal comprehensive health data to output the patient's health assessment result. The patient's health assessment result is used to assist doctors in health diagnosis. The doctor inputs the diagnosis result into the computer (32), and the diagnosis result is transmitted to the intelligent terminal (11) through the server-side wireless transmission module (31) and the terminal wireless transmission module (2) for display.

8. An intelligent auxiliary diagnosis decision control method based on AI chip, characterized by: The intelligent terminal (11) of the intelligent auxiliary diagnosis and decision system based on the AI ​​chip according to claim 7 is used for control, comprising the following steps: Step 1, the patient operates the touch display screen (117) to select an auxiliary diagnosis option; Step 2, the processor (116) retrieves health data collection guidance video data, the health data collection guidance video data including a welcome interface video segment, a test data entry guidance video segment, and a question-and-answer form health self-report guidance video segment; Step 3, the processor (116) controls the touch display screen (117) to first play the welcome interface video segment, and then play the examination data entry guidance video segment, the examination data entry guidance video segment provides examination data entry options, the examination data entry options include a shooting entry option, a local file entry option and an examination data entry completion option, when the patient selects the shooting entry option, the camera module (111) is called to shoot, when the patient selects the local file entry option, the memory (113) is scanned to provide the patient with a picture or video selection; Step 4, when the patient selects the option of completing the entry of examination data, the processor (116) controls the touch screen (117) to play a video segment in which the patient self-reports his / her health condition in the form of a question and answer. After each question in the video segment in which the patient self-reports his / her health condition in the form of a question and answer is completed, the microphone module (112) is called to collect the patient's self-reported voice data; Step 5, when the self-reporting voice data of the patient in the last item of the question-and-answer form health condition self-reporting guidance video segment is collected, the original data collection of the health status information description of the smart terminal (11) is completed; Step 6, the edge data storage (121) communicates and interacts with the storage (113) to exchange the original data describing the health status information, and the AI ​​chip (122) performs edge health status diagnosis and analysis on the original data describing the health status information based on the deep learning model. The analysis results of the edge health status diagnosis and analysis include multiple diagnosis results and the probabilities corresponding to the diagnosis results, and the edge health status diagnosis and analysis results are displayed through the touch display screen (117). The patient, based on the basic cognition of his own situation, eliminates several diagnosis results that are determined to be inconsistent with his own health status by operating the touch display screen (117); In step 7, the processor (116) packages the remaining diagnosis results and adds a timestamp and patient information tag data to form a comprehensive health data packet of the patient terminal, and transmits the comprehensive health data packet of the patient terminal to the server-side medical large model analysis module (3) through the terminal wireless transmission module (2).

9. The intelligent auxiliary diagnosis decision control method based on AI chip according to claim 8 is characterized in that: In step 6, the AI ​​chip (122) performs edge health status diagnosis analysis on the raw data describing the health status information based on the deep learning model, including the following sub-steps: Step 61, digitizing the patient health examination data and the patient health-related data; Step 62, normalizing or standardizing the data; Step 63, merging the feature vectors of the health condition question and answer data corresponding to different types of examination list data and patient health-related data into a comprehensive feature vector X; Step 64, input the comprehensive feature vector X into the neural network of the AI ​​chip (122), and the output layer of the neural network of the AI ​​chip (122) uses a softmax activation function to output multiple diagnosis results and corresponding probabilities, and selects the top ten diagnosis results and corresponding probabilities for display.

10. The AI ​​chip-based intelligent auxiliary diagnosis decision control method according to claim 9, characterized in that: For each health status information describing the original data i and each diagnosis result category j, the softmax function is calculated as follows: Where P(y i |X i ) indicates that in the feature vector X i The probability of diagnosis result category j under the input, e is the base of natural logarithm, Z ij is the eigenvector X i The raw score corresponding to diagnosis category j, Denotes the feature vector X i Corresponding to the raw score for diagnosis result category k in the neural network, m is the total number of diagnosis result categories.

Images