Remote cerebral apoplexy management method and system based on artificial intelligence

Abstract

The invention provides a remote cerebral apoplexy management method and system based on artificial intelligence, and belongs to the field of artificial intelligence. The problem of low cerebral apoplexy diagnosis efficiency is solved; the method specifically comprises the following steps: acquiring physiological data, pathological data and pharmacological data by a medical record acquisition module; the medical record analysis module integrates the physiological data, the pathological data and the pharmacological data to obtain a comprehensive physiological parameter group and a comprehensive pharmacological matching group of the cerebral apoplexy, and constructs a picture matching model of the cerebral apoplexy; the remote pre-diagnosis module obtains to-be-detected data; the possibility that the patient suffers from the cerebral apoplexy is judged by combining the to-be-detected data, the comprehensive physiological parameter set and the image matching model; if the possibility is high, the lesion symptoms of the patient are summarized, and pre-diagnosis data are obtained; if the possibility is low, processing is not carried out; the registration appointment module registers the patient according to the pre-diagnosis data; according to the method, the stroke pathogenesis characteristics are extracted through the medical records of the target region, whether the patient is ill or not is judged according to the stroke pathogenesis characteristics, and the stroke diagnosis efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and particularly to a remote stroke management method and system based on artificial intelligence. Background Art

[0002] The existing stroke management systems and methods have the following deficiencies: High data dependence and low data utilization rate: Most of the existing systems or methods rely on AI technology, namely AI systems; the performance of AI systems or methods highly depends on the quality and quantity of medical record information (i.e., training information); if there are deviations in the medical record information or errors in the extraction of medical record information, the performance of the AI system will be severely affected; in addition, the deviation of the algorithm or model of the AI system will lead to inaccurate data analysis and processing results, which is not conducive to the diagnosis and treatment of stroke patients.

[0003] Low popularity: The popularization and promotion of the existing systems are narrow; on the one hand, the development and maintenance of the existing systems require high-cost investment, including the costs of hardware equipment, software development, personnel training, etc.; on the other hand, the existing operations and feedback are not user-friendly to medical workers, resulting in uneven application effects of the existing systems.

[0004] Lack of processing objects: Although the existing systems can optimize their performance through learning and adaptation, they perform poorly in the tasks of stroke analysis and diagnosis; in stroke management, doctors need to make flexible decisions according to the specific conditions and changes in the condition of patients, while the existing systems often can only process specific populations according to preset algorithms and models, and the decisions for specific populations are referential; the existing systems often lack the ability to handle non-specific populations, and the decisions for non-specific populations are less referential compared to those for specific populations. Summary of the Invention

[0005] Aiming at the deficiencies of the existing technology, the purpose of the present invention is to provide a remote stroke management method and system based on artificial intelligence, aiming to solve the problem of low stroke diagnosis efficiency.

[0006] To achieve the above purpose, the present invention is realized through the following technical solutions: A remote stroke management system based on artificial intelligence includes: Medical record acquisition module: used to acquire all medical record information of the target area as reference medical records; extract the age, gender, blood pressure, blood sugar, blood lipid, and blood oxygen saturation corresponding to all stroke patients from the reference medical records to obtain physiological data; Extract the onset symptoms and lesion pictures corresponding to all stroke patients to obtain pathological data; Extract the drug names corresponding to all stroke patients to obtain pharmacological data; Medical record analysis module: used to analyze and integrate physiological data, pathological data, and pharmacological data to obtain a comprehensive set of physiological parameters and a comprehensive pharmacological matching group for stroke; summarize the lesion images of all patients in the reference medical records as a picture set, and based on the AlexNet model, construct a picture matching model for stroke; Remote pre-diagnosis module: used to obtain the current physiological data and lesion images of the patient as the data to be examined; combine the data to be examined, the comprehensive set of physiological parameters, and the picture matching model to judge the possibility that the patient has stroke; if the possibility is high, summarize the patient's lesion symptoms and issue a prescription for the patient according to the comprehensive pharmacological matching group to obtain pre-diagnosis data; if the possibility is low, do not process; Registration and appointment module: used to register for the patient according to the pre-diagnosis data; continuously monitor the stroke incidence risk in the target area.

[0007] Furthermore, the working process of the medical record analysis module is as follows: Process A1: Count the total number of reference medical records, denoted as cn; Denote the minimum age of stroke patients corresponding to the reference medical records as agl, and the maximum age of stroke patients as agm; Count the number of reference medical records with the patient's age being agl, denoted as pn （agl） ; Count the number of reference medical records with the patient's age being (agl + 1), denoted as pn （agl＋1） ; And so on, count the number of reference medical records with the patient's age being agm, denoted as pn （agm） ; Process A2: Use the patient's age as the abscissa and the number of patients as the ordinate to construct an age - number curve, denoted as curve Qyc; define relationship a: [lim (Qyc`(x) / Qyc`(x - 1)) → 1] ∩ [lim (Qyc`(x) / Qyc`(x + 1)) → 1]; where x represents the abscissa of curve Qyc, Qyc`(x) represents the slope of curve Qyc at abscissa x, Qyc`(x - 1) represents the slope of curve Qyc at abscissa (x - 1), and Qyc`(x + 1) represents the slope of curve Qyc at abscissa (x + 1); Process A3: In curve Qyc, find the first abscissa that satisfies relationship a and denote it as aga; Taking aga as the starting point, find the first abscissa in curve Qyc that does not satisfy relationship a and denote it as agb; take [aga, agb] as the high-incidence age range of stroke; Take [agl, aga) as interval 1, [aga, agb) as interval 2, and [agb, agm] as interval 3; Calculate the total number of reference medical records of patients in age range 1, denoted as cna; Calculate the total number of reference medical records of patients in age range 2, denoted as cnb; Calculate the total number of reference medical records of patients in age range 3, denoted as cnc; Calculate the proportion coefficient of patients in range 1, denoted as pr1, pr1 = cna / cn; Calculate the proportion coefficient of patients in range 2, denoted as pr2, pr2 = cnb / cn; Calculate the proportion coefficient of patients in range 3, denoted as pr3, pr3 = cnc / cn; Process A4: Aggregate the data in Processes A1 to A3, and calculate the comprehensive physiological parameter group of stroke; Process A5: Count the total number of types of all drug names in the reference cases, denoted as dn; Analyze the corresponding relationship between each drug and the onset symptoms based on the reference cases to obtain the comprehensive pharmacological matching group of stroke; Process A6: Aggregate the lesion pictures of all stroke patients in the reference cases as a picture set; Based on the AlexNet model as the basic model, construct a picture matching model for stroke.

[0008] Furthermore, the specific process of the said Process A4 is as follows: Process A41: Sequentially count and calculate the average blood pressure value of male patients in age range 1, denoted as bpm1; average blood sugar value, denoted as bsm1; average blood lipid value, denoted as blm1; average blood oxygen saturation value, denoted as exm1; Sequentially count and calculate the average blood pressure value of male patients in age range 2, denoted as bpm2; average blood sugar value, denoted as bsm2; average blood lipid value, denoted as blm2; average blood oxygen saturation value, denoted as exm2; Sequentially count and calculate the average blood pressure value of male patients in age range 3, denoted as bpm3; average blood sugar value, denoted as bsm3; average blood lipid value, denoted as blm3; average blood oxygen saturation value, denoted as exm3; Process A42: Calculate the weighted blood pressure value of male patients, denoted as bpm; bpm = pr1 × bpm1 + pr2 × bpm2 + pr3 × bpm3; Weighted blood sugar value, denoted as bsm; bsm = pr1 × bsm1 + pr2 × bsm2 + pr3 × bsm3; Weighted blood lipid value, denoted as blm; blm = pr1 × blm1 + pr2 × blm2 + pr3 × blm3; Weighted blood oxygen saturation value, denoted as oxm; oxm = pr1 × oxm1 + pr2 × oxm2 + pr3 × oxm3; Take bpm, bsm, blm, and oxm as the comprehensive male physiological parameter group; Process A43: Sequentially count and calculate the average blood pressure of female patients in interval 1, denoted as bpw1; the average blood glucose, denoted as bsw1; the average blood lipid, denoted as blw1; and the average blood oxygen saturation, denoted as exw1; Sequentially count and calculate the average blood pressure of female patients in interval 2, denoted as bpw2; the average blood glucose, denoted as bsw2; the average blood lipid, denoted as blw2; and the average blood oxygen saturation, denoted as exw2; Sequentially count and calculate the average blood pressure of female patients in interval 3, denoted as bpw3; the average blood glucose, denoted as bsw3; the average blood lipid, denoted as blw3; and the average blood oxygen saturation, denoted as exw3; Process A44: Calculate the weighted blood pressure value of female patients, denoted as bpw; bpw = pr1 × bpw1 + pr2 × bpw2 + pr3 × bpw3; The weighted blood glucose value, denoted as bsw; bsw = pr1 × bsw1 + pr2 × bsw2 + pr3 × bsw3; The weighted blood lipid value, denoted as blw; blw = pr1 × blw1 + pr2 × blw2 + pr3 × blw3; The weighted blood oxygen saturation value, denoted as oxw; oxw = pr1 × oxw1 + pr2 × oxw2 + pr3 × oxw3; Take bpw, bsw, blw, and oxw as the comprehensive female physiological parameter group; Process A45: Take the comprehensive male physiological parameter group and the comprehensive female physiological parameter group as the comprehensive physiological parameter group for stroke.

[0009] Furthermore, the specific process of Process A5 is as follows: Process A51: Define the expression b: dm(i) （j） ; where both i and j are positive numbers, the value range of i is: 1 to cn, and the value range of j is: 1 to dn; dm(i) （j） represents the usage times of the jth drug in the ith reference medical record; Use the expression b to count the usage times of the 1st to the dnth drugs in the 1st reference medical record, and obtain dm(1) （1） 、dm(1) （2） ~dm(1)（dn） ; where, dm(1) （1） represents the number of times the first drug is used in the first reference medical record; dm(1) （2） represents the number of times the second drug is used in the first reference medical record; and so on, dm(1) （dn） represents the number of times the dn-th drug is used in the first reference medical record; Use the expression b to count the number of times the first to dn-th drugs are used in the second reference medical record, obtaining dm(2) （1） , dm(2) （2） ~dm(2) （dn） ; where, dm(2) （1） represents the number of times the first drug is used in the second reference medical record; dm(2) （2） represents the number of times the second drug is used in the second reference medical record; and so on, dm(2) （dn） represents the number of times the dn-th drug is used in the second reference medical record; And so on, use the expression b to count the number of times the first to dn-th drugs are used in the cn-th reference medical record, obtaining dm(cn) （1） , dm(cn) （2） ~dm(cn) （dn） ; where, dm(cn) （1） represents the number of times the first drug is used in the cn-th reference medical record; dm(cn) （2） represents the number of times the second drug is used in the cn-th reference medical record; and so on, dm(cn) （dn） represents the number of times the dn-th drug is used in the cn-th reference medical record; Process A52: Calculate the total number of times all drugs are used in the reference medical records, denoted as dmm; Calculate the usage frequency of the first drug, denoted as mp(1); The calculation formula for mp(1) is as follows: ; where, i is a positive number, and the value range of i is 1~cn; dm(i) （1） represents the number of times the first drug is used in the i-th reference medical record; Calculate the usage frequency of the second drug, denoted as mp(2); The calculation formula for mp(2) is as follows: ; where, dm(i) （2） represents the number of times the second drug is used in the i-th reference medical record; And so on, calculate the usage frequency of the dn-th drug, denoted as mp(dn); The calculation formula for mp(dn) is as follows: ; where, dm(i) （dn）Denote the number of times the dn-th drug is used in the i-th reference medical record; Process A53: Summarize and process all the onset symptoms in the reference cases to obtain the main symptoms and secondary symptoms of stroke; Process A54: Integrate the usage frequencies of the 1st to dn-th drugs according to the main symptoms and secondary symptoms of stroke to obtain the comprehensive pharmacological matching group of stroke.

[0010] Further, the specific process of Process A53 is as follows: Process A531: Count the number of types of all the onset symptoms in the reference cases, denoted as c; Count the number of all the onset symptoms in the reference cases, denoted as oc; Process A532: Count the number of the 1st onset symptom in the reference cases, denoted as n1; Count the number of the 2nd onset symptom in the reference cases, denoted as n2; And so on, count the number of the c-th onset symptom in the reference cases, denoted as nc; Process A533: Calculate the TF-IDF values corresponding to the 1st to c-th onset symptoms to obtain TF-IDF(n1) ~ TF-IDF(nc); Calculate the TF-IDF value of the 1st onset symptom to obtain TF-IDF(n1); Calculate the inverse document frequency IDF(n1) of the 1st onset symptom. The calculation formula of IDF(n1) is as follows: ; Calculate the corresponding TF-IDF(n1) value of the 1st onset symptom. The calculation formula of TF-IDF(n1) is as follows: ; Calculate the TF-IDF value of the 2nd onset symptom to obtain TF-IDF(n2); Calculate the inverse document frequency IDF(n2) of the 2nd onset symptom. The calculation formula of IDF(n2) is as follows: ; Calculate the corresponding TF-IDF(n2) value of the 2nd onset symptom. The calculation formula of TF-IDF(n2) is as follows: ; And so on, calculate the TF-IDF value of the c-th onset symptom to obtain TF-IDF(nc); Calculate the inverse document frequency IDF(nc) of the c-th onset symptom. The calculation formula of IDF(nc) is as follows: ; Calculate the TF-IDF (nc) value corresponding to the c-th disease symptom. The calculation formula of TF-IDF (nc) is as follows: ; Process A534: Arrange the 1st to c-th disease symptoms in descending order of TF-IDF (n1) to TF-IDF (nc) to obtain sequence g; Using the Symptom database as the reference library, map the disease symptoms in sequence g to the reference library using the word vector model GloVe to obtain the high-dimensional vectors of each disease symptom, denoted as vg(1) to vg(c); Summarize vg(1) to vg(c) as set g`.

[0011] Furthermore, the subsequent process of the process A534 is as follows: Process A535: Define the relational expression c: vg(r) = λ × vg(a) + (1 - λ) × vg(b); vg(r), vg(a), and vg(b) satisfy: {[vg(r) & vg(a) & vg(b)] ∈ set g`} ∩ [vg(r) ≠ vg(a) ≠ vg(b)]; where r is a positive number, and the value range of r is: 1 to c; vg(r) represents the high-dimensional vector of the r-th disease symptom in set g`; λ represents any rational number in the interval (0, 1); a and b are positive numbers, and the value ranges of both a and b are: 1 to c; vg(a) represents the high-dimensional vector of the a-th disease symptom in set g`; vg(a) is default to increase sequentially backward from the first item vg(1) in set g`; vg(b) represents the high-dimensional vector of the b-th disease symptom in set g`; vg(b) is default to decrease sequentially forward from the last item vg(c) in set g`; Process A536: In set g`, summarize the high-dimensional vectors that satisfy the relational expression c as the main high-dimensional vectors; summarize the high-dimensional vectors that do not satisfy the relational expression c as the secondary high-dimensional vectors; Summarize the disease symptoms corresponding to the main high-dimensional vectors as the main symptoms of stroke, and count the number of main symptoms, denoted as mn; Summarize the disease symptoms corresponding to the secondary high-dimensional vectors as the secondary symptoms of stroke, and count the number of secondary symptoms, denoted as ln.

[0012] Furthermore, the specific process of the process A54 is as follows: Process A541: Determine the matching drugs for the main symptoms of stroke; Count the number of the 1st to the mn-th main symptoms in the reference cases, denoted as cm(1), cm(2) ~ cm(mn); where cm(1) represents the number of the 1st main symptom in the reference cases; cm(2) represents the number of the 2nd main symptom in the reference cases; and so on, cm(mn) represents the number of the mn-th main symptom in the reference cases; Process A542: Define the relational expression d: lim[(cm(p) × mp(q)) / oc] → 1; Where both p and q are positive numbers, the value range of p is: 1 ~ mn, and the value range of q is: 1 ~ dn; cm(p) represents the number of the p-th main symptom in the reference cases; mp(q) represents the usage frequency of the q-th drug; Process A543: Determine the matching drugs for the 1st main symptom; Substitute cm(1) and mp(1) ~ mp(dn) into the relational expression d, and extract the drugs that satisfy the relational expression d from the 1st to the dn-th drugs as alternative drugs; Compare the usage frequencies of the alternative drugs, and select the alternative drug with the highest usage frequency as the matching drug for the 1st disease; Process A544: Repeat the same process of determining the matching drugs for the 1st main symptom to determine the matching drugs for the 2nd to the mn-th main symptoms; Process A545: Repeat the same process of determining the matching drugs for the main symptoms to determine the matching drugs for the secondary symptoms of stroke; Summarize the matching drugs for the main symptoms and secondary symptoms of stroke as the comprehensive pharmacological matching group of stroke.

[0013] Furthermore, the specific process of the Process A6 is as follows: Process A61: Based on the AlexNet model as the basic model, adjust the initial structure of the AlexNet model to obtain the original model; Adjust the first layer; The convolution kernel size of the convolutional layer conv1: 5×5, the stride: 1, and the number of channels: 64; The convolution kernel size of the pooling layer pool1: 3×3, the stride: 2; The scaling factor of the LRN layer norm1 is 0.001 / 9, and the exponential term is 0.75; Process A62: Adjust the second layer; The convolution kernel size of the convolutional layer conv2: 5×5, the stride: 1, and the number of channels: 128; The convolution kernel size of the pooling layer pool2: 3×3, the stride: 2; The scaling factor of the LRN layer norm2 is 0.001 / 9, and the exponential term is 0.75; Process A63: Adjust the third layer; The convolution kernel size of the convolutional layer conv3: 5×5, stride: 1, number of channels: 64; The convolution kernel size of the pooling layer pool3: 3×3, stride: 2; Process A64: Adjust the fourth layer; The size of the weight matrix of the fully connected layer local4: 1024×384, with an initial value of 0.04; Adjust the fifth layer; The size of the weight matrix of the fully connected layer local5: 384×192, with an initial value of 0.1; The size of the weight matrix of the Softmax layer: 192×1016, with an initial value of 0.005; Process A65: Use the cross-entropy loss function as the loss function of the original model; Use the gradient descent optimization algorithm as the optimization algorithm of the original model; Use the L1 norm as the regularization term of the original model; Process A66: Back up the picture set to obtain picture set A and picture set B; Divide picture set A evenly into picture set A1 and picture set A2; Import the ImageNet picture library; use picture set A1 as the training set of the original model, use the ImageNet picture library and picture set A2 as the test set of the original model, use picture set B as the validation set of the original model, and train the original model until the loss function of the original model is minimized and each picture in picture set B has been output at least once to obtain the picture matching model for stroke.

[0014] Furthermore, the working process of the remote pre-diagnosis module is as follows: Process B1: Obtain the current physiological data, lesion symptoms, and lesion pictures of the patient as the data to be examined; The current physiological data of the patient includes the patient's age, gender, blood pressure, blood sugar, blood lipid, and blood oxygen saturation; Process B2: Denote the age of the patient in the data to be examined as ago, the blood pressure as bpo, the blood sugar as bso, the blood lipid as blo, and the blood oxygen saturation as oxo; Judge the size relationship between the patient's age and [aga, agb] in the multiple-occurrence age range to determine the age influence coefficient of the patient, denoted as bag; If ago < aga, the calculation formula for bag is bag = 1 - ago / aga; If ago ∈ [aga, agb], the calculation formula for bag is bag = 1; If ago > agb, the calculation formula for bag is bag = 1 + bab / ago; Process B3: Determine the patient's gender and calculate the primary incidence rate of stroke in the patient, denoted as fm; If the patient is male, the calculation formula for fm is as follows: fm = bag × [(bpo - bpm) / bpm + (bso - bsm) / bsm + (blo - blm) / blm + (oxo - oxm) / oxm] / 4; If the patient is female, the calculation formula for fm is as follows: fm = bag × [(bpo - bpw) / bpw + (bso - bsw) / bsw + (blo - blw) / blw + (oxo - oxw) / oxw] / 4; Process B4: Determine whether fm is greater than or equal to 0.5; If fm is greater than or equal to 0.5, it indicates that the patient has a high possibility of suffering from stroke, and select a drug that matches the lesion symptoms in the comprehensive pharmacological matching group of stroke as the patient's prescription; If fm is less than 0.5, analyze the secondary incidence rate of the patient's stroke based on the lesion picture, denoted as ftm, and enter Process B5; Process B5: Use the lesion picture of the patient as the input of the picture matching model, and summarize the output of the picture matching model as the comparison picture; Calculate the similarity between the lesion picture and the comparison picture as the value of the secondary incidence rate ftm; Process B6: Summarize the analysis results of Processes B1 to B5 as the pre-diagnosis data of the patient.

[0015] Furthermore, the specific process of Process B5 is as follows: Process B51: Use the cvtColor function in the OpenCV library (OpenCV is an open-source computer vision and machine learning software library) to obtain the grayscale value of the lesion picture and the grayscale value of the comparison picture; Process B52: Calculate the average value of the grayscale values of the lesion picture, denoted as ax, and the standard deviation, denoted as bx; Calculate the average value of the grayscale values of the comparison picture, denoted as ay, and the standard deviation, denoted as by; Calculate the covariance cxy of the grayscale values of the lesion picture and the comparison picture; Process B53: Calculate the brightness similarity between the lesion picture and the comparison picture, denoted as ll; the calculation formula for ll is as follows: ll = (2 × ax × ay + C1) / (ax 2+ ay 2 + C1); where C1 represents the brightness coefficient, and the calculation formula of C1 is as follows: ; where n represents the number of bits of the image; Process B54: Calculate the contrast similarity between the lesion picture and the comparison picture, denoted as cc; the calculation formula of cc is as follows: cc = (2 × bx × by + C2) / (bx 2 + by 2 + C2); where C2 represents the contrast coefficient, and the calculation formula of C2 is as follows: C2 = z × C1; where z is an additional parameter; Process B55: Calculate the structural similarity between the lesion picture and the comparison picture, denoted as ss; the calculation formula of ss is as follows: ss = (cxy + C3) / (bx × by + C3); where C3 represents the structure coefficient, and the calculation formula of C2 is: C3 = C2 / 2; Process B56: Calculate the value of ftm: ftm = ll × cc × ss; Judge whether ftm is greater than or equal to 0.5; If ftm is greater than or equal to 0.5, it indicates that the patient has a high possibility of suffering from stroke, and select the drug that matches the lesion symptoms in the comprehensive pharmacological matching group of stroke as the patient's prescription; If ftm is less than 0.5, it indicates that the patient has a low possibility of suffering from stroke and no treatment is required.

[0016] An artificial intelligence-based remote stroke management method includes: Step S1: Obtain all the medical record information of the target area as the reference medical records; extract the age, gender, blood pressure, blood sugar, blood lipid, and blood oxygen saturation corresponding to all stroke patients in the reference medical records to obtain physiological data; Extract the onset symptoms and lesion pictures corresponding to all stroke patients to obtain pathological data; Extract the drug names corresponding to all stroke patients to obtain pharmacological data; Step S2: Analyze and integrate the physiological data, pathological data, and pharmacological data to obtain the comprehensive physiological parameter group and comprehensive pharmacological matching group of stroke; summarize the lesion pictures of all patients in the reference medical records as the picture set, and build a picture matching model of stroke based on the AlexNet model; Step S3: Obtain the patient's current physiological data and pathological images as the data to be examined; combine the data to be examined, the comprehensive physiological parameter group, and the image matching model to judge the possibility that the patient has a stroke; if the possibility is high, summarize the patient's pathological symptoms, and issue a prescription for the patient according to the comprehensive pharmacological matching group to obtain the preliminary diagnosis data; if the possibility is low, do not process. Step S4: Register for the patient according to the preliminary diagnosis data; continuously monitor the stroke incidence risk in the target area.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: Improve the diagnosis efficiency and accuracy: The present invention uses technologies such as image processing, natural language processing, and data mining to extract and analyze a large number of stroke medical records, and can automatically identify the pathological areas of patients, providing faster and more accurate diagnosis results than traditional methods, and shortening the diagnosis time.

[0018] Dynamic risk screening: Based on the patient's medical history, physiological indicators, and drug data, the present invention conducts dynamic risk assessment. Through machine learning algorithms, AI can continuously learn and optimize the risk assessment model, improving the sensitivity and specificity of screening; helping to discover potential populations of stroke and reducing the occurrence risk of stroke.

[0019] Improve the medical efficiency: By comparing and analyzing the patient's current various physiological data with the medical records of patients with stroke, the present invention evaluates the probability of the patient suffering from stroke, the medication, and the dosage of the medication, reducing the workload of medical staff; at the same time, the present invention can process and analyze existing medical records, extract the pathological information and pathological rules of stroke in the target area, providing rich data resources for medical workers and promoting the research progress in the field of stroke prevention and treatment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] By reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings, other features, objectives, and advantages of the present invention will become more apparent: Figure 1 It is a schematic diagram of the system of the present invention; Figure 2 It is a schematic diagram of the method of the present invention; Figure 3 It is a schematic diagram of the model structure of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0022] For the specific first embodiment, please refer to Figure 1, A remote stroke management system based on artificial intelligence includes: a medical record acquisition module, a medical record analysis module, a remote pre-diagnosis module, a registration and appointment module, a database, and a server; among them, the medical record acquisition module, the medical record analysis module, the remote pre-diagnosis module, and the registration and appointment module are respectively connected to the database and the server.

[0023] It should be noted that the "target area" in the present invention refers to: the municipal area where remote stroke management is carried out using the present invention (a remote stroke management system and method based on artificial intelligence); Medical record acquisition module: used to acquire the medical record information of all (stroke patients) in the target area as reference medical records; extract the age, gender, blood pressure, blood sugar, blood lipid, and blood oxygen saturation corresponding to all stroke patients from the reference medical records to obtain the (stroke) physiological data; Extract the onset symptoms and lesion pictures corresponding to all stroke patients from the reference medical records to obtain the (stroke) pathological data; Extract the drug names corresponding to all stroke patients from the reference medical records to obtain the (stroke) pharmacological data; It should be noted that in the process of the "medical record acquisition module" in the present invention acquiring the "stroke patient medical record information", for the same patient, the present invention acquires the most recent stroke medical record information of this patient.

[0024] Medical record analysis module: used to analyze and integrate the (stroke) physiological data, (stroke) pathological data, and (stroke) pharmacological data to obtain the comprehensive physiological parameter group and comprehensive pharmacological matching group of stroke; Summarize the lesion pictures of all (stroke) patients in the reference medical records as a picture set, and based on the AlexNet model, construct a picture matching model for stroke; Process A: The working process of the medical record analysis module is as follows: Process A1: Count the total number of reference medical records, denoted as cn; Denote the minimum age of the stroke patients corresponding to the reference medical records as agl, and the maximum age of the stroke patients as agm; Count the number of reference medical records with the patient age of agl, denoted as pn （agl） ; Count the number of reference medical records with the patient age of (agl + 1), denoted as pn （agl＋1） ; Count the number of reference medical records with the patient age of (agl + 2), denoted as pn （agl＋2） ; And so on, count the number of reference medical records with the patient age of agm, denoted as pn （agm） ; Process A2: Using the age of (stroke) patients as the abscissa and the number of (stroke) patients as the ordinate, construct an age - number curve, denoted as curve Qyc; Define relation a: [lim (Qyc`(x) / Qyc`(x - 1)) → 1] ∩ [lim (Qyc`(x) / Qyc`(x + 1)) → 1]; where x represents the abscissa of curve Qyc, Qyc`(x) represents the slope at abscissa x on curve Qyc, Qyc`(x - 1) represents the slope at abscissa (x - 1) on curve Qyc, and Qyc`(x + 1) represents the slope at abscissa (x + 1) on curve Qyc; Process A3: In curve Qyc, find the first abscissa that satisfies relation a, denoted as aga; Taking aga as the starting point, in curve Qyc, find the first abscissa that does not satisfy relation a, denoted as agb; Take [aga, agb] as the age range with high incidence of stroke; Take [agl, aga) as interval 1, [aga, agb) as interval 2, and [agb, agm] as interval 3; Calculate the total number of reference medical records of patients in interval 1 of age, denoted as cna; Calculate the total number of reference medical records of patients in interval 2 of age, denoted as cnb; Calculate the total number of reference medical records of patients in interval 3 of age, denoted as cnc; Calculate the proportion coefficient of patients in interval 1, denoted as pr1, pr1 = cna / cn; Calculate the proportion coefficient of patients in interval 2, denoted as pr2, pr2 = cnb / cn; Calculate the proportion coefficient of patients in interval 3, denoted as pr3, pr3 = cnc / cn; Process A4: Summarize the data in Processes A1 - A3, and calculate the comprehensive physiological parameter group of stroke; Process A41: Sequentially count and calculate the average blood pressure of male (stroke) patients in interval 1 of age, denoted as bpm1; The average blood sugar, denoted as bsm1; The average blood lipid, denoted as blm1; The average blood oxygen saturation, denoted as exm1; Sequentially count and calculate the average blood pressure of male (stroke) patients in interval 2 of age, denoted as bpm2; The average blood sugar, denoted as bsm2; The average blood lipid, denoted as blm2; The average blood oxygen saturation, denoted as exm2; Sequentially count and calculate the average blood pressure of male (stroke) patients in interval 3 of age, denoted as bpm3; The average blood sugar, denoted as bsm3; The average blood lipid, denoted as blm3; The average blood oxygen saturation, denoted as exm3; Process A42: Calculate the weighted blood pressure value for male (stroke) patients, denoted as bpm; bpm = pr1 × bpm1 + pr2 × bpm2 + pr3 × bpm3; Weighted blood glucose value, denoted as bsm; bsm = pr1 × bsm1 + pr2 × bsm2 + pr3 × bsm3; Weighted blood lipid value, denoted as blm; blm = pr1 × blm1 + pr2 × blm2 + pr3 × blm3; Weighted blood oxygen saturation value, denoted as oxm; oxm = pr1 × oxm1 + pr2 × oxm2 + pr3 × oxm3; Take bpm, bsm, blm, and oxm as the comprehensive male physiological parameter group (for stroke); Process A43: Sequentially count and calculate the average blood pressure value, denoted as bpw1, for female (stroke) patients in age range 1; average blood glucose value, denoted as bsw1; average blood lipid value, denoted as blw1; average blood oxygen saturation value, denoted as exw1; Sequentially count and calculate the average blood pressure value, denoted as bpw2, for female (stroke) patients in age range 2; average blood glucose value, denoted as bsw2; average blood lipid value, denoted as blw2; average blood oxygen saturation value, denoted as exw2; Sequentially count and calculate the average blood pressure value, denoted as bpw3, for female (stroke) patients in age range 3; average blood glucose value, denoted as bsw3; average blood lipid value, denoted as blw3; average blood oxygen saturation value, denoted as exw3; Process A44: Calculate the weighted blood pressure value for female (stroke) patients, denoted as bpw; bpw = pr1 × bpw1 + pr2 × bpw2 + pr3 × bpw3; Weighted blood glucose value, denoted as bsw; bsw = pr1 × bsw1 + pr2 × bsw2 + pr3 × bsw3; Weighted blood lipid value, denoted as blw; blw = pr1 × blw1 + pr2 × blw2 + pr3 × blw3; Weighted blood oxygen saturation value, denoted as oxw; oxw = pr1 × oxw1 + pr2 × oxw2 + pr3 × oxw3; Take bpw, bsw, blw, and oxw as the comprehensive female physiological parameter group (for stroke); Process A45: Combine the comprehensive male physiological parameter group (for stroke) and the comprehensive female physiological parameter group (for stroke) as the comprehensive physiological parameter group for stroke; Process A5: Count the number of types of all drug names in the reference cases, denoted as dn; Analyze the corresponding relationship between each drug and the onset symptoms based on the reference cases to obtain the comprehensive pharmacological matching group for stroke; Process A51: Define the expression b: dm(i) （j） ; where both i and j are positive numbers, the value range of i is: 1 to cn, and the value range of j is: 1 to dn; dm(i) （j） represents the usage times of the jth drug in the ith reference medical record; (the minimum value of dm(i) （j） is 0, and the maximum value is 1) Use the expression b to count the usage times of the 1st to dnth drugs in the 1st reference medical record to obtain dm(1) （1） , dm(1) （2） ~dm(1) （dn） ; where dm(1) （1） represents the usage times of the 1st drug in the 1st reference medical record; dm(1) （2） represents the usage times of the 2nd drug in the 1st reference medical record; and so on, dm(1) （dn） represents the usage times of the dnth drug in the 1st reference medical record; Use the expression b to count the usage times of the 1st to dnth drugs in the 2nd reference medical record to obtain dm(2) （1） , dm(2) （2） ~dm(2) （dn） ; where dm(2) （1） represents the usage times of the 1st drug in the 2nd reference medical record; dm(2) （2） represents the usage times of the 2nd drug in the 2nd reference medical record; and so on, dm(2) （dn） represents the usage times of the dnth drug in the 2nd reference medical record; And so on, use the expression b to count the usage times of the 1st to dnth drugs in the cnth reference medical record to obtain dm(cn) （1） , dm(cn) （2） ~dm(cn) （dn） ; where dm(cn) （1） represents the usage times of the 1st drug in the cnth reference medical record; dm(cn) （2） represents the usage times of the 2nd drug in the cnth reference medical record; and so on, dm(cn) （dn） represents the usage times of the dnth drug in the cnth reference medical record; Process A52: Calculate the total number of times all drugs in the reference medical records are used (i.e., the sum of dm(1) （1） ~dm(cn) （dn） , and denote it as dmm; Calculate the usage frequency of the first drug, denoted as mp(1); the calculation formula of mp(1) is as follows: ; where i is a positive number, and the value range of i is 1~cn; dm(i) （1） represents the number of times the first drug is used in the i-th reference medical record; Calculate the usage frequency of the second drug, denoted as mp(2); the calculation formula of mp(2) is as follows: ; where dm(i) （2） represents the number of times the second drug is used in the i-th reference medical record; And so on, calculate the usage frequency of the dn-th drug, denoted as mp(dn); the calculation formula of mp(dn) is as follows: ; where dm(i) （dn） represents the number of times the dn-th drug is used in the i-th reference medical record; Process A53: Summarize and process the onset symptoms of all (stroke patients) in the reference cases to obtain the main symptoms and secondary symptoms of stroke; Process A531: Count the number of types of all onset symptoms (word segments) in the reference cases (i.e., the number of different onset symptom word segments in the reference cases), denoted as c; Count the number of all onset symptoms (word segments) in the reference cases (i.e., the number of all onset symptom word segments in the reference cases), denoted as oc; Process A532: Count the number of the first onset symptom in the reference cases, denoted as n1; Count the number of the second onset symptom in the reference cases, denoted as n2; And so on, count the number of the c-th onset symptom in the reference cases, denoted as nc; Process A533: Calculate the TF-IDF values corresponding to the first to c-th onset symptoms to obtain TF-IDF(n1)~TF-IDF(nc); Calculate the TF-IDF value of the first onset symptom to obtain TF-IDF(n1); Calculate the inverse document frequency IDF(n1) of the first onset symptom, and the calculation formula of IDF(n1) is as follows: ; Calculate the TF-IDF(n1) value corresponding to the first onset symptom, and the calculation formula of TF-IDF(n1) is as follows: ; Calculate the TF-IDF value of the second type of onset symptom to obtain TF-IDF(n2); Calculate the inverse document frequency IDF(n2) of the second type of onset symptom. The calculation formula of IDF(n2) is as follows: ; Calculate the TF-IDF(n2) value corresponding to the second type of onset symptom. The calculation formula of TF-IDF(n2) is as follows: ; And so on, calculate the TF-IDF value of the c-th type of onset symptom to obtain TF-IDF(nc); Calculate the inverse document frequency IDF(nc) of the c-th type of onset symptom. The calculation formula of IDF(nc) is as follows: ; Calculate the TF-IDF(nc) value corresponding to the c-th type of onset symptom. The calculation formula of TF-IDF(nc) is as follows: ; Process A534: Arrange the first to c-th onset symptoms in descending order of TF-IDF(n1) to TF-IDF(nc) to obtain the sequence g of (onset symptom phrases); Using the Symptom database (the Symptom database is a systematic database containing more than 100,000 disease names and their related symptom descriptions) as a reference library, use the word vector model GloVe to map the onset symptoms (word segments) in sequence g to the reference library to obtain the high-dimensional vectors of each onset symptom (word segment), denoted as vg(1) to vg(c); Summarize vg(1) to vg(c) as set g`; Process A535: Define the relationship formula c: vg(r) = λ × vg(a) + (1 - λ) × vg(b); vg(r), vg(a), and vg(b) satisfy: {[vg(r) & vg(a) & vg(b)] ∈ set g`} ∩ [vg(r) ≠ vg(a) ≠ vg(b)]; Among them, r is a positive number, and the value range of r is: 1 to c; vg(r) represents the high-dimensional vector of the r-th type of onset symptom (word segment) in set g`; λ represents any rational number in the interval (0, 1); a and b are positive numbers, and the value ranges of both a and b are: 1 to c; vg(a) represents the high-dimensional vector of the a-th type of onset symptom (word segment) in set g`; vg(a) is default to increase sequentially backward from the first item vg(1) in set g`; vg(b) represents the high-dimensional vector of the b-th onset symptom (word segment) in the set g`; by default, vg(b) decreases successively backward from the last item vg(c) in the set g`. Process A536: In the set g`, summarize the high-dimensional vectors that satisfy the relational expression c as the main high-dimensional vectors; summarize the high-dimensional vectors that do not satisfy the relational expression c as the secondary high-dimensional vectors. Summarize the onset symptoms (word segments) corresponding to the main high-dimensional vectors as the main symptoms (word segments) of stroke, and count the number of the main symptoms (word segments), denoted as mn. Summarize the onset symptoms (word segments) corresponding to the secondary high-dimensional vectors as the secondary symptoms (word segments) of stroke, and count the number of the secondary symptoms (word segments), denoted as ln. Process A54: Integrate the usage frequencies of the 1st to the dn-th drugs according to the main symptoms (word segments) and secondary symptoms (word segments) of stroke to obtain the comprehensive pharmacological matching group of stroke. Process A541: Determine the matching drugs for the main symptoms (word segments) of stroke. Count the number of the 1st to the mn-th main symptoms (word segments) in the reference cases, denoted as cm(1), cm(2) ~ cm(mn); where cm(1) represents the number of the 1st main symptom (word segment) in the reference cases; cm(2) represents the number of the 2nd main symptom (word segment) in the reference cases; and so on, cm(mn) represents the number of the mn-th main symptom (word segment) in the reference cases. Process A542: Define the relational expression d: lim[(cm(p) × mp(q)) / oc] → 1. Where both p and q are positive numbers, the value range of p is: 1 ~ mn, and the value range of q is: 1 ~ dn. cm(p) represents the number of the p-th main symptom (word segment) in the reference cases. mp(q) represents the usage frequency of the q-th drug. Process A543: Determine the matching drugs for the 1st main symptom (word segment). Substitute cm(1) and mp(1) ~ mp(dn) into the relational expression d, and extract the drugs that satisfy the relational expression d from the 1st to the dn-th drugs as the alternative drugs. Compare the usage frequencies of the alternative drugs, and select the alternative drug with the highest usage frequency as the matching drug for the 1st disease. Process A544: Repeat the same process as determining the matching drugs for the 1st main symptom (word segment) (i.e., Process A543) to determine the matching drugs for the 2nd to the mn-th main symptoms (word segments). Process A545: Repeatedly determine the matching drugs for the secondary symptoms (word segments) of stroke by the same process of determining the matching drugs for the primary symptoms (word segments) (i.e., Process A541 to Process A544). Summarize the matching drugs for the primary symptoms (word segments) and the matching drugs for the secondary symptoms (word segments) of stroke as the comprehensive pharmacological matching group for stroke. Process A6: Summarize the lesion images of all stroke patients in the reference cases as a picture set. Process A61: Refer to Figure 3 and, based on the AlexNet model as the basic model, adjust the initial structure of the AlexNet model to obtain the original model. Adjust the first layer (of the AlexNet model). The convolution kernel size of the convolutional layer conv1: 5×5, stride: 1, number of channels: 64. The convolution kernel size of the pooling layer pool1: 3×3, stride: 2. The scaling factor of the LRN layer (local response normalization layer) norm1: 0.001 / 9, exponent term: 0.75. Process A62: Adjust the second layer (of the AlexNet model). The convolution kernel size of the convolutional layer conv2: 5×5, stride: 1, number of channels: 128. The convolution kernel size of the pooling layer pool2: 3×3, stride: 2. The scaling factor of the LRN layer (local response normalization layer) norm2: 0.001 / 9, exponent term: 0.75. Process A63: Adjust the third layer (of the AlexNet model). The convolution kernel size of the convolutional layer conv3: 5×5, stride: 1, number of channels: 64. The convolution kernel size of the pooling layer pool3: 3×3, stride: 2. Process A64: Adjust the fourth layer (of the AlexNet model). The weight matrix size of the fully connected layer local4: 1024×384, initial value: 0.04. Adjust the fifth layer (of the AlexNet model). The weight matrix size of the fully connected layer local5: 384×192, initial value: 0.1. The weight matrix size of the Softmax layer: 192×1016, initial value: 0.005. Process A65: Use the cross-entropy loss function as the loss function of the original model. Use the gradient descent optimization algorithm as the optimization algorithm of the original model. Use the L1 norm as the regularization term for the original model; Process A66: Backup the picture set to obtain picture set A and picture set B; Divide picture set A evenly into picture set A1 and picture set A2; Import the ImageNet picture library (ImageNet is a large visual database for research on picture recognition software, containing more than 14 million labeled pictures); use picture set A1 as the training set for the original model, use the ImageNet picture library and picture set A2 as the test set for the original model, and use picture set B as the validation set for the original model. Train the original model until the loss function of the original model is minimized and each picture in picture set B has been output at least once to obtain the picture matching model for stroke.

[0025] Remote pre-diagnosis module: Used to obtain the current physiological data and lesion pictures of the patient as the data to be examined; combine the data to be examined, the comprehensive physiological parameter group, and the picture matching model to judge the possibility that the patient has a stroke; if the possibility is high, summarize the patient's lesion symptoms and issue a prescription for the patient according to the comprehensive pharmacology matching group to obtain the pre-diagnosis data; if the possibility is low, do not process; Process B: The working process of the remote pre-diagnosis module is as follows: Process B1: Obtain the current physiological data, lesion symptoms, and lesion pictures of the patient as the data to be examined; The current physiological data of the patient includes the patient's age, gender, blood pressure, blood sugar, blood lipid, and blood oxygen saturation; Process B2: Denote the patient's age in the data to be examined as ago, blood pressure as bpo, blood sugar as bso, blood lipid as blo, and blood oxygen saturation as oxo; Judge the size relationship between the patient's age and [aga, agb] (the multi-occurrence age range of stroke) to determine the age influence coefficient of the patient, denoted as bag; If ago < aga, the calculation formula for bag is bag = 1 - ago / aga; If ago ∈ [aga, agb], the calculation formula for bag is bag = 1; If ago > agb, the calculation formula for bag is bag = 1 + bab / ago; Process B3: Judge the patient's gender and calculate the one-time incidence rate of stroke for the patient, denoted as fm; If the patient is male, the calculation formula for fm is as follows: fm = bag × [(bpo - bpm) / bpm + (bso - bsm) / bsm + (blo - blm) / blm + (oxo - oxm) / oxm] / 4; If the patient is female, the calculation formula for fm is as follows: fm = bag × [((bpo - bpw) / bpw + (bso - bsw) / bsw + (blo - blw) / blw + (oxo - oxw) / oxw) / 4]; Process B4: Determine whether fm is greater than or equal to 0.5; If fm is greater than or equal to 0.5, it indicates that the patient has a high probability of suffering from stroke. Select a drug that matches the lesion symptoms from the comprehensive pharmacological matching group for stroke as the patient's prescription; If fm is less than 0.5, analyze the secondary incidence rate of the patient's stroke based on the lesion pictures, denoted as ftm, and enter Process B5; Process B5: Use the patient's lesion pictures as the input of the picture matching model, and summarize the output of the picture matching model as the comparison pictures; Calculate the similarity between the lesion pictures and the comparison pictures as the value of the secondary incidence rate ftm; Process B51: Use the cvtColor function in the OpenCV library (OpenCV is an open-source computer vision and machine learning software library) to obtain the grayscale values of the lesion pictures and the comparison pictures; Process B52: Calculate the average value of the grayscale values of the lesion pictures, denoted as ax, and the standard deviation, denoted as bx; Calculate the average value of the grayscale values of the comparison pictures, denoted as ay, and the standard deviation, denoted as by; Calculate the covariance cxy of the grayscale values of the lesion pictures and the comparison pictures; Process B53: Calculate the brightness similarity between the lesion pictures and the comparison pictures, denoted as ll; The calculation formula for ll is as follows: ll = (2 × ax × ay + C1) / (ax 2 + ay 2 + C1); where, C1 represents the brightness coefficient, and the calculation formula for C1 is as follows: ; where, n represents the number of image bits (i.e., the dynamic comparison range of the picture pixel values), and n generally takes the value of 8 (users or relevant technical personnel can adjust the value of n according to actual needs); Process B54: Calculate the contrast similarity between the lesion pictures and the comparison pictures, denoted as cc; The calculation formula for cc is as follows: cc = (2 × bx × by + C2) / (bx 2 + by 2 + C2); where, C2 represents the contrast coefficient, and the calculation formula for C2 is as follows: C2 = z × C1; where z is an additional parameter, z represents a positive integer in the range [1, 9], and z generally takes the value of 3 (users or relevant technical personnel can adjust the value of z according to actual needs); Process B55: Calculate the structural similarity between the lesion image and the comparison image, denoted as ss; the calculation formula for ss is as follows: ss = (cxy + C3) / (bx × by + C3); where C3 represents the structural coefficient, and the calculation formula for C2 is: C3 = C2 / 2; Process B56: Calculate the value of ftm: ftm = ll × cc × ss; Determine whether ftm is greater than or equal to 0.5; If ftm is greater than or equal to 0.5, it indicates that the patient has a high probability of suffering from stroke, and select a drug that matches the lesion symptoms from the comprehensive pharmacological matching group for stroke as the patient's prescription; If ftm is less than 0.5, it indicates that the patient has a low probability of suffering from stroke and do not take any action; Process B6: Summarize the analysis results of Processes B1 to B5 as the preliminary diagnosis data of the patient.

[0026] Registration and appointment module: used to register patients based on the preliminary diagnosis data; continuously monitor the stroke incidence risk (i.e., the stroke incidence rate) in the target area.

[0027] Example Two Please refer to Figure 2 , a remote stroke management method based on artificial intelligence includes: Step S1: Obtain the medical record information of all (stroke patients) in the target area as the reference medical records; extract the age, gender, blood pressure, blood sugar, blood lipid, and blood oxygen saturation corresponding to all stroke patients from the reference medical records to obtain the (stroke) physiological data; Extract the onset symptoms and lesion images corresponding to all stroke patients from the (reference medical records) to obtain the (stroke) pathological data; Extract the drug names corresponding to all stroke patients from the (reference medical records) to obtain the (stroke) pharmacological data; Step S2: Analyze and integrate the (stroke) physiological data, (stroke) pathological data, and (stroke) pharmacological data to obtain the comprehensive physiological parameter group and comprehensive pharmacological matching group for stroke; summarize the lesion images of all (stroke) patients in the reference medical records as a picture set, and build a picture matching model for stroke based on the AlexNet model; Step S3: Obtain the current physiological data and lesion pictures of the patient as the data to be examined; combine the data to be examined, the comprehensive physiological parameter group, and the picture matching model to judge the possibility that the patient has a stroke; if the possibility is high, summarize the patient's lesion symptoms and issue a prescription for the patient according to the comprehensive pharmacological matching group to obtain the preliminary diagnosis data; if the possibility is low, do not process. Step S4: Register for the patient according to the preliminary diagnosis data; continuously monitor the stroke incidence risk (i.e., the stroke incidence rate) in the target area.

[0028] All the above formulas are calculated by taking the numerical values without dimensions. The formula is obtained by collecting a large amount of data for software simulation to get a formula closest to the actual situation. The preset parameters in the formula are set by those skilled in the art according to the actual situation. For example, if there are weight coefficients and proportionality coefficients, the values set are for quantifying each parameter to obtain a specific numerical value for subsequent comparison. Regarding the magnitudes of the weight coefficients and proportionality coefficients, as long as they do not affect the proportional relationship between the parameters and the quantified numerical values, it is acceptable.

[0029] Finally, it should be noted that the above embodiments are only specific embodiments of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that any person skilled in the technical field of the present invention can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A remote stroke management system based on artificial intelligence, characterized in that, The system includes: Medical record acquisition module: used to acquire all medical record information of the target area as reference medical records; extract the age, gender, blood pressure, blood sugar, blood lipid, and blood oxygen saturation corresponding to all stroke patients from the reference medical records to obtain physiological data; Extract the onset symptoms and lesion pictures corresponding to all stroke patients to obtain pathological data; Extract the drug names corresponding to all stroke patients to obtain pharmacological data; Medical record analysis module: used to analyze and integrate physiological data, pathological data, and pharmacological data to obtain a comprehensive physiological parameter group and a comprehensive pharmacological matching group for stroke; summarize the lesion pictures of all patients in the reference medical records as a picture set, and build a picture matching model for stroke based on the AlexNet model; Remote pre-diagnosis module: used to acquire the current physiological data and lesion pictures of the patient as data to be examined; combine the data to be examined, the comprehensive physiological parameter group, and the picture matching model to judge the possibility that the patient has a stroke; if the possibility is high, summarize the lesion symptoms of the patient, and issue a prescription for the patient according to the comprehensive pharmacological matching group to obtain pre-diagnosis data; if the possibility is low, do not process; Registration and appointment module: used to register the patient according to the pre-diagnosis data; continuously monitor the stroke incidence risk in the target area.

2. The remote stroke management system based on artificial intelligence according to claim 1, characterized in that, The working process of the medical record analysis module is as follows: Process A1: Count the total number of reference medical records cn; Record the minimum age of stroke patients as agl and the maximum age as agm; Count the number of reference medical records pn with the patient age of agl （agl） ; the number of reference medical records pn with the patient age of agm （agm） ; Process A2: Construct an age - number curve with patient age as the abscissa and the number of patients as the ordinate, denoted as curve Qyc; Find the age range [aga, agb] with the highest incidence of stroke according to the slope change on curve Qyc; Process A3: Take [agl, aga) as interval 1, [aga, agb) as interval 2, and [agb, agm] as interval 3; Calculate the total number of reference medical records cna of patients in interval 1; the total number of reference medical records cnb in interval 2; the total number of reference medical records cnc in interval 3; Calculate the proportion coefficient pr1 in interval 1, pr2 in interval 2, and pr3 in interval 3 of the patients; Process A4: Calculate the comprehensive physiological parameter group for stroke; Process A5: Count the number of types of all drug names dn in the reference cases; Analyze the corresponding relationship between each drug and the onset symptoms according to the reference cases to obtain the comprehensive pharmacological matching group for stroke; Process A6: Summarize the lesion pictures of all stroke patients in the reference cases as a picture set; build a picture matching model for stroke based on the AlexNet model as the basic model, and enter the remote pre-diagnosis module.

3. The remote stroke management system based on artificial intelligence according to claim 2, characterized in that, The specific process of Process A4 is as follows: Process A41: Count and calculate the average blood pressure bpm1, average blood sugar bsm1, average blood lipid blm1, and average blood oxygen saturation exm1 of male patients in interval 1; The average blood pressure bpm2, average blood sugar bsm2, average blood lipid blm2, and average blood oxygen saturation exm2 in interval 2; The average blood pressure in interval 3, bpm3; the average blood glucose, bsm3; the average blood lipid, blm3; the average blood oxygen saturation, exm3; Procedure A42: Calculate the weighted blood pressure value bpm for male patients: bpm = pr1 × bpm1 + pr2 × bpm2 + pr3 × bpm3; The weighted blood glucose value bsm: bsm = pr1 × bsm1 + pr2 × bsm2 + pr3 × bsm3; The weighted blood lipid value blm: blm = pr1 × blm1 + pr2 × blm2 + pr3 × blm3; The weighted blood oxygen saturation value oxm: oxm = pr1 × oxm1 + pr2 × oxm2 + pr3 × oxm3; Take bpm, bsm, blm, and oxm as the comprehensive male physiological parameter group; Procedure A43: Calculate the average blood pressure bpw1, the average blood glucose bsw1, the average blood lipid blw1, and the average blood oxygen saturation exw1 for female patients in interval 1; The average blood pressure bpw2, the average blood glucose bsw2, the average blood lipid blw2, and the average blood oxygen saturation exw2 in interval 2; The average blood pressure bpw3, the average blood glucose bsw3, the average blood lipid blw3, and the average blood oxygen saturation exw3 in interval 3; Procedure A44: Calculate the weighted blood pressure value bpw, the weighted blood glucose value bsw, the weighted blood lipid value blw, and the weighted blood oxygen saturation value oxw for female patients, and take them as the comprehensive female physiological parameter group; Procedure A45: Take the comprehensive male physiological parameter group and the comprehensive female physiological parameter group as the comprehensive physiological parameter group for stroke.

4. The remote stroke management system based on artificial intelligence according to claim 2, characterized in that The specific procedure of the said Procedure A5 is as follows: Process A51: Count the usage times dm(1) of the 1st to the dnth drugs in the 1st reference medical record （1） ~dm(1) （dn） ; the usage times dm(cn) of the 1st to the dnth drugs in the cnth reference medical record （1） ~dm(cn) （dn） ; Procedure A52: Calculate the total usage times dmm of all drugs in the reference medical record; Calculate the usage frequency mp(1) of the first drug: ; where i is a positive number, dm(i) （1） represents the number of times the first drug in the i-th reference medical record is used; Calculate the usage frequency mp(dn) of the dnth drug; Procedure A53: Summarize and process all the onset symptoms in the reference case to obtain the main symptoms and secondary symptoms of stroke; Procedure A54: Integrate the usage frequencies of the first to dnth drugs according to the main symptoms and secondary symptoms of stroke to obtain the comprehensive pharmacological matching group for stroke.

5. The remote stroke management system based on artificial intelligence according to claim 4, characterized in that, The specific procedure of the said Procedure A53 is as follows: Procedure A531: Count the number of types c of all onset symptoms in the reference case, and the number of all onset symptoms is denoted as oc; Procedure A532: Count the number n1 of the first onset symptom in the reference case; Similarly, the number nc of the cth onset symptom; Procedure A533: Calculate the inverse document frequency IDF(n1) of the first onset symptom; ； Calculate the TF-IDF(n1) corresponding to the first onset symptom; ； Similarly, calculate the inverse document frequency IDF(nc) of the cth onset symptom; calculate the TF-IDF(nc) corresponding to the cth onset symptom; Procedure A534: Arrange the first to cth onset symptoms in descending order of TF-IDF(n1)~TF-IDF(nc) to obtain the sequence g; Using the Symptom database as the reference library, the disease onset symptoms in sequence g are mapped to the reference library using the word vector model GloVe to obtain the high-dimensional vectors of each disease onset symptom, denoted as vg(1) to vg(c). Summarize vg(1) to vg(c) as set g`.

6. The remote stroke management system based on artificial intelligence according to claim 5, wherein The subsequent process of process A534 is as follows: Process A535: Define the relationship c: vg(r) = λ × vg(a) + (1 - λ) × vg(b); vg(r), vg(a), and vg(b) satisfy: {[vg(r) & vg(a) & vg(b)] ∈ set g`} ∩ [vg(r) ≠ vg(a) ≠ vg(b)]; where vg(r) represents the high-dimensional vector of the i-th disease onset symptom in set g`; λ represents any rational number in the interval (0, 1); a and b are positive numbers; vg(a) represents the high-dimensional vector of the a-th disease onset symptom in set g`; vg(b) represents the high-dimensional vector of the b-th disease onset symptom in set g`; Process A536: In set g`, summarize the high-dimensional vectors that satisfy relationship c as the main high-dimensional vectors; summarize the high-dimensional vectors that do not satisfy relationship c as the secondary high-dimensional vectors; Summarize the disease onset symptoms corresponding to the main high-dimensional vectors as the main symptoms of stroke, and count the number mn of the main symptoms; Summarize the disease onset symptoms corresponding to the secondary high-dimensional vectors as the secondary symptoms of stroke, and count the number ln of the secondary symptoms.

7. An artificial intelligence-based remote stroke management system according to claim 4, characterized in that, The specific process of process A54 is as follows: Process A541: Determine the matching drugs for the main symptoms of stroke; Count the number cm(1) to cm(mn) of the 1st to the mn-th main symptoms in the reference cases; Process A542: Define the relationship d: lim[(cm(p) × mp(q)) / oc] → 1; cm(p) represents the number of the p-th main symptom in the reference cases; mp(q) represents the usage frequency of the q-th drug; Process A543: Determine the matching drugs for the 1st main symptom; Substitute cm(1) and mp(1) to mp(dn) into relationship d, and extract the drugs that satisfy relationship d from the 1st to the dn-th drugs as the alternative drugs; Compare the usage frequencies of the alternative drugs, and select the alternative drug with the highest usage frequency as the matching drug for the 1st disease; Process A544: Determine the matching drugs for the 2nd to the mn-th main symptoms; Determine the matching drugs for the secondary symptoms of stroke; Summarize the matching drugs for the main symptoms and the matching drugs for the secondary symptoms of stroke as the comprehensive pharmacological matching group of stroke.

8. An artificial intelligence-based remote stroke management system according to claim 2, characterized in that, The working process of the remote pre-diagnosis module is as follows: Process B1: Obtain the data to be examined; Denote the age of the patient in the data to be examined as ago, blood pressure as bpo, blood sugar as bso, blood lipid as blo, and blood oxygen saturation as oxo; Process B2: Determine the age influence coefficient bag of the patient according to the patient's age; If ago < aga, then bag is: bag = 1 - ago / aga; If ago ∈ [aga, agb], then bag is: bag = 1; If ago > agb, then bag is: bag = 1 + bab / ago; Process B3: Determine the gender of the patient, and calculate the primary incidence rate of stroke in the patient according to the patient's gender, denoted as fm; If the patient is male, the calculation formula for fm is: fm = bag × [(bpo - bpm) / bpm + (bso - bsm) / bsm + (blo - blm) / blm + (oxo - oxm) / oxm] / 4; If the patient is female, the calculation formula for fm is: fm = bag × [(bpo - bpw) / bpw + (bso - bsw) / bsw + (blo - blw) / blw + (oxo - oxw) / oxw] / 4; Process B4: Determine whether fm is greater than or equal to 0.5; If fm is greater than or equal to 0.5, it indicates that the patient has a high possibility of suffering from stroke, and select a drug that matches the lesion symptoms from the comprehensive pharmacological matching group for stroke as the patient's prescription; If fm is less than 0.5, analyze the secondary incidence rate of the patient's stroke based on the lesion pictures; Process B5: Use the patient's lesion pictures as the input of the picture matching model, and summarize the output of the picture matching model as the comparison picture; Calculate the similarity between the lesion picture and the comparison picture as the secondary incidence rate ftm; Summarize the analysis results of Process B1 to Process B5 as the preliminary diagnosis data of the patient.

9. The remote stroke management system based on artificial intelligence according to claim 8, wherein, The specific process of the said Process B5 is as follows: Process B51: Use the cvtColor function in the OpenCV library to obtain the grayscale value of the lesion picture and the grayscale value of the comparison picture; Process B52: Calculate the average value ax and the standard deviation bx of the grayscale value of the lesion picture; Calculate the average value ay and the standard deviation by of the grayscale value of the comparison picture; Calculate the covariance cxy of the grayscale values of the lesion picture and the comparison picture; Process B53: Calculate the brightness similarity ll between the lesion picture and the comparison picture; ll = (2 × ax × ay + C1) / (ax 2 + ay 2 + C1); where C1 represents the brightness coefficient, and the calculation formula of C1 is: ; n represents the number of bits of the image; Process B54: Calculate the contrast similarity cc between the lesion picture and the comparison picture; cc = (2 × bx × by + C2) / (bx 2 + by 2 + C2); C2 represents the contrast coefficient, and the calculation formula of C2 is: C2 = z × C1; z is an additional parameter; Process B55: Calculate the structure similarity ss between the lesion picture and the comparison picture; ss = (cxy + C3) / (bx × by + C3); C3 represents the structure coefficient, and the calculation formula for C2 is: C3 = C2 / 2; Process B56: Calculate the value of ftm: ftm = ll × cc × ss; Determine whether ftm is greater than or equal to 0.5; If ftm is greater than or equal to 0.5, it indicates that the patient has a high possibility of suffering from stroke, and select a drug that matches the lesion symptoms from the comprehensive pharmacological matching group for stroke as the patient's prescription; If ftm is less than 0.5, do not process.

10. A remote stroke management method based on artificial intelligence, applicable to a remote stroke management system according to any one of claims 1-9, characterized in that, The said method includes: Step S1: Obtain all the medical record information of the target area as the reference medical records; extract the age, gender, blood pressure, blood sugar, blood lipid and blood oxygen saturation corresponding to all stroke patients in the reference medical records to obtain physiological data; Extract the onset symptoms and lesion pictures corresponding to all stroke patients to obtain pathological data; Extract the drug names corresponding to all stroke patients to obtain pharmacological data; Step S2: Analyze and integrate the physiological data, pathological data, and pharmacological data to obtain the comprehensive physiological parameter group and comprehensive pharmacological matching group for stroke; summarize the lesion pictures of all patients in the reference medical records as a picture set, and based on the AlexNet model, construct a picture matching model for stroke; Step S3: Obtain the current physiological data and lesion pictures of the patient as the data to be examined; combine the data to be examined, the comprehensive physiological parameter group, and the picture matching model to judge the possibility that the patient has stroke; if the possibility is high, summarize the lesion symptoms of the patient, and issue a prescription for the patient according to the comprehensive pharmacological matching group to obtain the preliminary diagnosis data; if the possibility is low, do not process; Step S4: Register the patient according to the preliminary diagnosis data; continuously monitor the stroke incidence risk in the target area.

Images