Physical examination management platform

Abstract

The invention discloses a physical examination management platform, and relates to the technical field of medical health management, and the physical examination management platform comprises the following steps: S1, a pre-examination stage; s2, a middle detection stage; and S3, a post-inspection stage. According to the physical examination management platform, AR action signs such as muscle tremor and swallowing actions are fused with traditional indexes such as family history and blood sugar, the limitation that in the prior art, only static data is relied on is broken through, the feature dimension is improved, the disease risk prediction accuracy is improved, the risk probability is updated immediately after a user supplements data, and the user experience is improved. Therefore, the depth of personalized physical examination service for each user is improved, and the sequence of physical examination items performed by the user is adjusted in real time on the physical examination day based on the position of the user in combination with the current progress of each item, the time required for one time and the moving path, so that the user is prevented from ineffective round-trip of frequently checking the progress of each item.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical health management, in particular to a health examination management platform. BACKGROUND

[0002] The health examination management platform is a digital system integrating medical information technology, data management technology and health management concept, which provides full-process examination services and health management for users. It breaks the limitations of information dispersion, complicated process and fragmented health management in traditional examination, and realizes the whole-cycle closed-loop management from pre-examination preparation, examination process optimization to post-examination health intervention and long-term follow-up through digital means.

[0003] The existing health examination management platform mainly relies on users to fill in static forms (such as age, gender, and medical history) and historical examination reports. Although it also uses questionnaires to infer disease risks for users and generate personalized examination items, the effective information collected by questionnaires is extremely limited, and the feedback information from non-professional users is not effective, resulting in insufficient depth of personalized examination items recommended by the existing health examination management platform, and high one-sidedness of disease risk assessment. SUMMARY

[0004] In view of the deficiencies of the prior art, the present application provides a health examination management platform to solve the problems raised in the background.

[0005] To achieve the above purpose, the present application realizes the following technical scheme: a health examination management platform comprising the following step process:

[0006] S1, pre-examination stage:

[0007] The user fills in the basic information and lifestyle on the APP and retrieves the past examination data, and the analysis system infers potential diseases based on the above data, and triggers the question and answer mechanism and AR action guidance. According to the user's reply content based on the disease and the analysis of the physical signs, the occurrence probability of the potential disease is analyzed, and the examination items are dynamically adjusted and reserved after the user's consent;

[0008] S2, examination stage:

[0009] After the examinee arrives at the hospital, the hospital background system predicts the time when the examinee can perform each item based on the current progress of each examination item and the time required for each item, and preliminarily arranges the items in order, and adjusts the preliminary arrangement based on the time required for the examinee to reach each item location, and then adjusts the adjusted item arrangement in real time based on the real-time changes of the current progress of each examination item, so that the examinee only needs to perform the examination in the order of the arranged items.

[0010] Further, in the step S1, the basic information includes, but is not limited to, the user's name, age, gender, family history, and lifestyle, including, but not limited to, diet, exercise, and sleep.

[0011] Further, in the step S1, the question and answer mechanism is based on the suspected potential disease and its characteristic manifestations to initiate inquiries to the user, and based on the user's answers to further infer the probability of the disease.

[0012] Further, in the step S1, the AR action guide is performed in cooperation with the question and answer mechanism, and the AR action guide is guided by an AR superimposed guidance layer such as a red outline line marking the shooting range, combined with edge computing to detect image quality such as definition and body position correctness in real time, to obtain the user's signs corresponding to the suspected disease.

[0013] Further, in the step S1, the calculation process of the probability of the potential disease is as follows:

[0014] Collect basic information: Through the user's medical health management APP, the demographic characteristics such as age, gender, height, weight, family history such as diabetes, cancer, wearable device data heart rate, sleep duration, lifestyle such as diet, exercise, and sleep are obtained;

[0015] Retrieve historical physical examination data: After user authorization, cross-institutional physical examination reports are called through a block link, and structured indicators such as blood pressure, blood glucose, blood lipids, and image conclusions such as "thyroid nodule TI-RADS 3" are analyzed;

[0016] Question and answer data: Based on the preliminary risk prediction, a progressive question chain is generated, such as for diabetes risk: "Have you been drinking and urinating more in the past half year?" "How often do you have symptoms?" ;

[0017] Sign data: Through the medical health management AR medical health management, the user is guided to complete specific actions such as thyroid region shooting and hand tremor video recording, and sign features such as nodule size and tremor frequency are extracted;

[0018] Missing value processing: Key indicators such as fasting blood glucose are prompted to be supplemented, and non-key indicators such as exercise frequency are filled with median values;

[0019] Standardization processing: Numerical data such as BMI is normalized to the 0-1 interval using Min-Max normalization, and categorical data such as "family history" is converted to binary labels;

[0020] Outlier correction: Extreme values such as systolic pressure > 200 mmHg are identified by the IQR method, and the user is prompted to confirm or replaced with the mean value of the same age group;

[0021] Extract general features: Calculate derived indicators such as BMI = weight ÷ height², smoking index = daily cigarette consumption × years, exercise compliance rate = actual exercise duration ÷ recommended duration;

[0022] Extract disease-specific features: Filter key factors by disease type, such as "fasting blood glucose, waist circumference, and frequency of polyuria symptoms" for diabetes; "gender, and the proportion of thyroid enlargement in AR images" for thyroid disease;

[0023] Extract interactive fusion features: Fuse question and answer data with physical data, such as "nighttime acid reflux frequency + degree of incoordination of throat muscles during swallowing movements" to generate a reflux risk index;

[0024] Rule engine preliminary screening: Set thresholds based on clinical guidelines, such as "fasting blood glucose ≥ 7.0 mmol / L + polydipsia and polyuria symptoms" → initial diabetes risk setting of 80%;

[0025] Multi-modal model actuarial: Input features into a Transformer fusion model, output probability values such as diabetes risk 72%, thyroid nodule risk 45%;

[0026] Real-time update: User supplements data such as new blood glucose values, after completing AR actions, immediately repeat the above process to update risk probabilities.

[0027] Further, disease risk classification: Divide into low risk <30%, medium risk 30%-70%, and high risk >70% based on probability values. Probabilities of diseases within the medium risk range can suggest users to undergo medical examinations.

[0028] Further, the health examination management platform further comprises the following step processes:

[0029] S3, post-examination phase: Integrate physical examination reports, genetic testing data, and wearable device information to generate a full-life-cycle health record, and based on a knowledge graph-based health risk assessment model, push personalized health recommendations.

[0030] Further, in step S3, personalized health recommendations include but are not limited to dietary adjustments and exercise plans.

[0031] The present application provides a health examination management platform, which has the following beneficial effects:

[0032] The health examination management platform fuses AR action signs such as muscle tremor and swallowing action with traditional indexes such as family history and blood sugar, breaks through the limitation of only relying on static data in the prior art, improves the feature dimension, and thus improves the disease risk prediction accuracy, and supports instant updating of the risk probability after the user supplements data, thereby improving the depth of personalized health examination services for each user, and adjusting the order of the user to perform the health examination items in real time based on the location where the user is located, the current progress of each item, the required time for each item and the moving path on the examination day, so as to avoid the invalid round trip of the user frequently going back and forth to check the progress of each item. BRIEF DESCRIPTION OF DRAWINGS

[0033] Figure 1 FIG. 1 is a schematic diagram of the step flow of the health examination management platform of the present application. DETAILED DESCRIPTION

[0034] The embodiments of the present application will be further described in detail below in combination with the drawings and examples. The following examples are used to illustrate the present application, but cannot be used to limit the scope of the present application.

[0035] As shown in FIG. 1, the present application provides a technical solution: a health examination management platform, comprising the following step flow: Figure 1 S1, pre-examination stage:

[0036] The user fills in the basic information and lifestyle on the APP and retrieves the past health examination data, the analysis system infers the potential disease based on the above data, triggers the question and answer mechanism and AR action guidance, analyzes the occurrence probability of the potential disease according to the reply content of the user based on the disease and the sign performance, and then dynamically adjusts the health examination items and makes project reservation after obtaining the consent of the user;

[0037] The basic information includes but is not limited to the user's name, age, gender, family history, and the lifestyle includes but is not limited to diet, exercise, and sleep. The question and answer mechanism is to ask the user based on the characteristic performance of the inferred potential disease, further infers the occurrence probability of the disease based on the user's answer, and the AR action guidance is performed together with the question and answer mechanism. The AR action guidance guides the shooting range through the AR superimposed guidance layer such as red contour line, and combines edge computing to detect the image quality such as definition and body position correctness in real time, to obtain the sign performance corresponding to the inferred disease of the user;

[0038] The AR action guidance is performed together with the question and answer mechanism, and the specific technical details are as follows:

[0039]

[0040] ​AR action guidance acquires user's signs corresponding to the presumed disease through AR superimposed guidance layers (such as red contour line marks the shooting range) combined with edge computing real-time detection of image quality (such as definition, body position correctness), wherein the specific generation method of the AR superimposed guidance layer is: based on the OpenPose human key point detection algorithm to identify the target part of the user (such as the thyroid area, the hand), a dynamic red contour line matching the contour of the target part is generated, and the size of the contour line is adjusted in real time according to the distance between the user and the device (the distance parameter is obtained through the device camera ranging module, and the linear scaling algorithm is used to adjust the pixel size of the contour line); the edge computing uses Canny edge detection operator (threshold value is minVal=50, maxVal=150) to extract the edge features of the target part in the image, combines with the SSIM (Structural Similarity Index) algorithm to detect the image definition (SSIM≥0.8 is determined as clear), and judges the body position correctness through the key point matching degree algorithm (calculates the Euclidean distance between the detected human key points and the standard body position key points, and the distance≤50 pixels is determined as the body position is correct), only when the definition and the body position meet the requirements, the sign data is collected;

[0041] For AR dynamic acquisition of hand tremor video (such as the action of holding a cup of water), the video acquisition parameters are set as frame rate 30fps, resolution 1920×1080 (1080P), and acquisition time 10 seconds (to ensure covering the complete tremor period); the specific steps of combining the Transformer model to analyze the tremor frequency and amplitude are as follows:

[0042] The collected video frames are preprocessed, and the hand movement area is extracted by inter-frame difference method to remove background interference;

[0043] The optical flow method (Farneback algorithm) is used to calculate the displacement vector of each frame of hand key points (such as fingertips, wrist joints);

[0044] The displacement vector sequence is input into the encoder (2 layers of encoder, 4 attention heads) of the Transformer model, and the self-attention mechanism is used to capture the time sequence correlation of the displacement;

[0045] The feature vector output by the model is extracted, the tremor frequency (the number of displacement direction changes per unit time, based on Fourier transform to convert the displacement sequence to frequency domain, and the peak frequency is the tremor frequency) and the tremor amplitude (the average value of the modulus of the displacement vector) are calculated, and finally the tremor characteristic parameters are output for disease risk assessment;

[0046] The calculation process of the probability of potential disease is as follows:

[0047] Collect basic information: Obtain demographic characteristics such as age, gender, height, weight, family history such as diabetes, cancer, wearable device data heart rate, sleep duration, lifestyle such as diet, exercise, sleep through user end medical health management APP medical health management;

[0048] Retrieve historical physical examination data: After user authorization, cross-institutional physical examination reports are retrieved through block chain interface, and structured indicators such as blood pressure, blood glucose, blood lipids and image conclusions such as "thyroid nodule TI-RADS 3" are analyzed;

[0049] Question and answer data: Based on preliminary risk prediction, generate progressive question chain such as for diabetes risk: "Have you been drinking and urinating more in the past half year?" "How often do you have symptoms?" ;

[0050] Sign data: Through medical health management AR medical health management, guide users to complete specific actions such as thyroid area shooting, hand tremor video recording, and extract sign features such as nodule size, tremor frequency;

[0051] Missing value processing: Key indicators such as fasting blood glucose prompt users to supplement, non-key indicators such as exercise frequency are filled with median;

[0052] Standardization: Numerical data such as BMI is normalized to 0-1 interval using Min-Max normalization, and categorical data such as "family history" is converted to binary label;

[0053] Outlier correction: Identify extreme values such as systolic pressure > 200 mmHg using IQR method, prompt user to confirm or replace with mean value of the same age group;

[0054] Extract general features: Calculate derived indicators such as BMI = weight ÷ height², smoking index = daily cigarette consumption × years, exercise compliance rate = actual exercise duration ÷ recommended duration;

[0055] Extract disease-specific features: Filter key factors according to disease type, such as diabetes-related "fasting blood glucose, waist circumference, and frequency of polyuria symptoms"; thyroid disease-related "gender, AR image thyroid enlargement ratio";

[0056] Extract interactive fusion features: Fuse question and answer data with sign data such as "nighttime acid reflux frequency + swallowing motion throat muscle incoordination degree" to generate reflux risk index;

[0057] Rule engine preliminary screening: Set threshold based on clinical guidelines such as "fasting blood glucose ≥7.0 mmol / L + polydipsia and polyuria symptoms" → diabetes risk initial setting 80%;

[0058] Multimodal model actuarial: input features into the Transformer fusion model, output probability values such as diabetes risk 72%, thyroid nodule risk 45%, specifically use a Transformer-based multimodal fusion model, input features include three types of structured and unstructured data, which are converted into vector form recognizable by the model after preprocessing:

[0059] Structured feature vector (dimension: 128): basic features (age, gender, BMI), biochemical indicators (fasting blood glucose, 2-hour postprandial blood glucose), behavioral characteristics (exercise compliance rate, smoking index);

[0060] Text feature vector (dimension: 256): extract semantic features from question and answer data using BERT model to generate symptom severity vector;

[0061] Image feature vector (dimension: 512): extract dynamic features from AR collected videos using 3D-CNN to generate sign dynamic vector;

[0062] The Transformer multimodal fusion model uses a 6-layer encoder structure, each layer contains 8 attention heads, and the hidden layer dimension is set to 1024; before feature splicing, first map the structured feature vector (128), text feature vector (256), and image feature vector (512) to 256 dimensions through a 1x1 convolutional layer, then use dimension splicing (concat) to form an initial fusion vector of 768 dimensions; modal identification uses one-hot encoding, structured features correspond to [1, 0, 0], text features correspond to [0, 1, 0], and image features correspond to [0, 0, 1], multiply the modal identification vector with each feature vector to achieve identification embedding; position encoding uses the sine-cosine position encoding formula:

[0063]

[0064]

[0065] where is the feature time sequence position (such as fasting blood glucose corresponding to , postprandial blood glucose corresponding to ), , is the dimension index, add the calculated position encoding vector to the feature vector after embedding the modal identification to complete the time sequence information injection, and the general calculation method of feature correlation weight is based on the self-attention score formula:

[0066]

[0067] where, is the query vector, is the key vector, respectively taken from different types of feature vectors (such as is the fasting blood glucose feature vector, is the polydipsia and polyuria symptom feature vector, is the dimension (256), the attention score calculated by the formula is the correlation weight between features, and after normalization by the softmax function, it is used for feature fusion weighting;

[0068] The Transformer self-attention mechanism is used to add modal identification to the three types of feature vectors, and the feature time sequence is distinguished by position coding. After calculating the correlation weight between features, the fused feature matrix is output. After being compressed to 2 dimensions by a 3-layer fully connected network, the probability distribution is calculated by the softmax function;

[0069] Real-time update: after the user supplements data such as new blood glucose values and completes AR actions, the above process is repeated in real time to update the risk probability. Disease risk classification: according to the probability value, it is divided into low risk < 30%, medium risk 30%-70%, and high risk > 70%. The disease probability in the medium risk range can suggest that the user undergo a physical examination;

[0070] S2, the detection stage:

[0071] After the examinee arrives at the hospital, the hospital background system predicts the time when he or she can perform each item based on the current progress of each item to be performed and the single required time, and preliminarily arranges the items in order. Then, based on the time required for the examinee to reach each item location from the current position, the preliminary arrangement is adjusted. Then, based on the real-time changes of the current progress of each examination item, the adjusted item arrangement is adjusted in real time, so that the examinee only needs to perform the examination in the order of the arranged items.

[0072] The hospital background system uses an improved Dijkstra algorithm combined with a dynamic greedy strategy to realize the arrangement and adjustment of examination items. The specific algorithm details are as follows:

[0073] Input parameter definition: let the set of examination items be The current progress of each item is , , which means that the item can start immediately, the single required time is , the current position of the examinee is , the location of each item is , , and the movement time matrix between locations is ,

[0074] Objective function:

[0075]

[0076]

[0077]

[0078] ;

[0079] Preliminary arrangement step: calculate the startable time of each item :

[0080]

[0081] Wherein, is the average waiting time of the item history, based on the past 3 months of hospital data statistics;

[0082] Take as the priority index, and preliminarily sort the items according to the index from small to large to obtain the initial sequence ;

[0083] Real-time adjustment step: update the progress and moving time matrix of each item every 2 minutes ; Based on the indoor positioning system of the hospital (such as UWB positioning, positioning accuracy ≤0.5 meters), the real-time position of the examinee and the congestion situation of the item location are obtained, and the real-time priority of each item in the current sequence is dynamically corrected ;

[0084] ; ;

[0085]

[0086] Adopting a greedy strategy to adjust the sequence: if there is an item ( not in the current execution position) of , then exchange and position, generate a new ;

[0087] S3, after the inspection stage: integrate the physical examination report, genetic testing data, and wearable device information to generate a full life cycle health record, and based on the knowledge graph health risk assessment model, push personalized health suggestions, including but not limited to diet adjustment, exercise plan.

[0088] Example demonstration (take diabetes risk assessment as an example to evaluate the probability of potential diseases)

[0089] Step 1: Data collection

[0090] ​Basic information: user age 55 years old, male, height 175 cm, weight 85 kg (BMI=27.8), father with diabetes;

[0091] Historical data: fasting blood glucose 6.5 mmol / L in the past 1 year, no medication;

[0092] Question and answer data: "more drinking and more urination for more than half a year, 3 times a week" "no intentional weight loss, weight loss 3 kg";

[0093] AR sign data: guide the user to take "postprandial 2-hour fingertip blood glucose monitoring video", extract blood glucose value 7.8 mmol / L;

[0094] Step 2: Preprocessing

[0095] No missing value, BMI normalized to 0.65, blood glucose value standardized to 0.72;

[0096] Step 3: Feature extraction

[0097] General features: BMI=27.8 (overweight), exercise compliance rate 0.3 (exercise 2 times a week, recommended 5 times);

[0098] Disease-specific features: fasting blood glucose 6.5 mmol / L, postprandial 2 hours 7.8 mmol / L, family history (yes), polyuria frequency 0.4 (3 times / 7 days per week);

[0099] Interaction features: "polyuria frequency x blood glucose fluctuation amplitude" = 0.4 x (7.8-6.5) = 0.52;

[0100] Step 4: Probability calculation

[0101] Rule engine preliminary screening: due to "fasting blood glucose 6.5 mmol / L (≥6.1) + family history + symptoms", initially set risk 60%;

[0102] Model actuarial: multi-modal model output probability 75%;

[0103] Specifically, a multi-modal fusion model based on Transformer is used, and the input features include three types of structured and unstructured data, which are converted into vector form recognizable by the model after preprocessing:

[0104] Structured feature vector (dimension: 128):

[0105] Basic features: age (55→ normalized 0.7), gender (male→ [1,0]), BMI (27.8→ normalized 0.65);

[0106] Biochemical indicators: fasting blood glucose (6.5 mmol / L → normalized 0.62), 2-hour postprandial blood glucose (7.8 mmol / L → normalized 0.72);

[0107] Behavioral characteristics: exercise compliance rate (0.3), smoking index (0, non-smoker).

[0108] Text feature vector (dimension: 256 dimensions):

[0109] For the question-answer data "more drinking and polyuria 3 times a week + weight loss 3 kg in the past half year", the semantic features are extracted by the BERT model to generate the "symptom severity vector" (0.82, high correlation).

[0110] Image feature vector (dimension: 512 dimensions):

[0111] For the "2-hour postprandial fingertip blood glucose monitoring video" collected by AR, the dynamic features (such as blood sampling operation standardization and blood glucose test paper color development speed) are extracted by 3D-CNN to generate "sign dynamic vector" (0.76, indicating abnormal blood glucose metabolism);

[0112] Use the Transformer self-attention mechanism to add "modality identifier" to the three types of feature vectors (structured = 001, text = 010, image = 100), and use position encoding to distinguish feature timing (such as the time sequence of "fasting blood glucose" and "postprandial blood glucose");

[0113] Calculate the correlation weight between features: for example, the attention weight between "fasting blood glucose (0.62)" and "polydipsia and polyuria symptoms (0.82)" is 0.85 (strong correlation), and the weight between "gender (male)" and "family history (yes)" is 0.72 (moderate correlation);

[0114] Output the fused feature matrix (dimension: 128+256+512=896), highlighting high-contribution features (such as the interaction term between blood glucose indicators and symptom text);

[0115] The fused feature matrix is compressed to 2 dimensions ("diabetes risk" "non-diabetes risk") by a 3-layer fully connected network, and the probability distribution is calculated by the softmax function:

[0116] Diabetes risk: exp(3.2) ÷ [exp(3.2) + exp(1.1)] = 0.75 → 75%;

[0117] Non-diabetes risk: 1-0.75=25%;

[0118] Where, exp(3.2) is a natural exponential function, which is used to convert the original score (positive or negative) output by the model into a non-negative value, which is convenient for subsequent probability calculation. The value of 3.2 is derived from the contribution of all input features, for example:

[0119] Positive features (increase the probability of disease): BMI overweight (+0.8), fasting blood glucose 6.5mmol / L (+1.2), family history (+0.6), polydipsia and polyuria symptoms (+0.7);

[0120] Negative features (reduce the probability of disease): blood glucose does not rise sharply without drug intervention (-0.1);

[0121] Finally, the model layer (such as the full connection layer) is weighted and summed to get 3.2;

[0122] The data source of exp(1.1) reflects the contribution of features supporting "not sick", for example: postprandial 2-hour blood glucose 7.8mmol / L does not reach the diabetes diagnosis threshold (+0.5), no severe complication symptoms (+0.6), etc. The weighted sum is 1.1;

[0123] exp(3.2)+exp(1.1): normalization term, the sum of the exponential values of all categories, which is used to normalize the non-negative value of the molecule to the interval of 0-1, to ensure that the final output is a result that meets the probability axiom (the sum is 1);

[0124] Step 5: Dynamic update

[0125] The user supplements "blood glucose has dropped to 5.8mmol / L after taking medicine for nearly 1 month", and the system recalculates in real time, and the risk is reduced to 45% (medium risk).

[0126] Based on the above, the application combines AR action signs such as muscle tremor and swallowing action with traditional indicators such as family history and blood glucose, breaks through the limitations of relying only on static data in the prior art, increases the feature dimension, and improves the accuracy of disease risk prediction. The risk probability is updated in real time after the user supplements the data, thereby improving the depth of personalized physical examination services for each user, and adjusting the order of the user to perform the physical examination items in real time based on the location of the user on the day of the physical examination combined with the current progress of each item, the required time for a single time and the moving path, to avoid the user's invalid round trip of frequently going back and forth to check the progress of each item.

[0127] Comparison of the effects of the platform and traditional technology:

[0128] Clinical verification test comparison table of 200 users, combined with traditional physical examination methods and platform test data, using randomized controlled trial design (control group 100 cases of traditional methods, experimental group 100 cases of platform methods), data verified by chi-square test (χ²) and t-test (P<0.05 is significant), the table below shows the comparison table of the platform and the traditional technology effect:

[0129] Disease type Traditional method accuracy Platform accuracy Sample size (positive / negative) P-value Diabetes risk assessment 60% (fasting blood glucose) 80% 100 cases (50 / 50) <0.001 Thyroid disease screening 55% (ultrasound palpation) 75% 50 cases (25 / 25) 0.005 Early symptoms of Parkinson's disease 40% (clinical scale) 60% 50 cases (25 / 25) 0.001

[0130] Data description:

[0131] Diabetes risk assessment

[0132] Traditional method: using fasting blood glucose test (accuracy rate 60%-70%), compared with the platform multi-modal data fusion scheme (AR hand tremor detection + glycosylated hemoglobin + family history), the accuracy rate is improved from 60% to 80%, with an increase of 33% (P<0.001);

[0133] Therefore, the platform collects hand tremor video through AR (such as the action of holding a cup of water), with video collection parameters of frame rate 30fps, resolution 1080P, collection time 10 seconds, combined with a 2-layer encoder, 4 attention heads of the Transformer model, displacement vectors are extracted by frame difference method and optical flow method, and tremor frequency and amplitude are calculated, which is more sensitive than traditional static blood glucose test;

[0134] Thyroid disease screening

[0135] Traditional method: relying on ultrasonic palpation (accuracy rate 40%-64%), while the platform uses AR neck image guidance (such as swallowing action shooting) combined with thyroid ultrasound data, AR guidance layer is generated based on OpenPose algorithm, edge calculation uses Canny operator, clarity determination SSIM≥0.8, body position determination key point Euclidean distance≤50 pixels, accuracy rate is improved from 55% to 75% (P=0.005);

[0136] Therefore, the platform uses AR red contour line to mark the thyroid boundary in real time, automatically identifies the size of nodules and blood flow signals, and reduces subjective errors compared with traditional manual ultrasound interpretation;

[0137] Early symptoms of Parkinson's disease

[0138] Traditional method: subjective assessment based on clinical scales (such as UPDRS) (accuracy rate 40%-50%), while the platform uses AR motion capture (such as finger-to-finger speed, gait stability) combined with wearable device data, accuracy rate is improved from 40% to 60% (P=0.001).

[0139] The embodiments of the present application are presented by way of example and description, and are not intended to be exhaustive or to limit the application to the form disclosed. Many modifications and variations will be apparent to those skilled in the art. Embodiments are chosen and described in order to best explain the principles of the application and its practical application, and to thereby enable others skilled in the art to best utilize the application in various embodiments and with various modifications as are suited to the particular use contemplated.

Claims

1. A health check-up management platform, characterized in that: The process includes the following steps: S1. Pre-inspection stage: Users fill in basic information and lifestyle details on the app and retrieve past physical examination data. The analysis system infers potential diseases based on the data and triggers a question-and-answer mechanism and AR action guidance. Based on the user's response to the disease and physical signs, the system analyzes the probability of the potential disease occurring, then dynamically adjusts the physical examination items and makes the items available for appointment after obtaining the user's consent. S2, Inspection Stage: Once the examinee arrives at the hospital, the hospital's backend system predicts the time required for each examination item based on the current progress and time needed for each item, and initially arranges them in chronological order. Then, it adjusts the initial arrangement based on the time required for the examinee to reach each item from their current location. Finally, it adjusts the arrangement in real time based on the real-time changes in the progress of each examination item, so that the examinee only needs to perform the examinations in the ordered sequence.

2. The health check-up management platform according to claim 1, characterized in that: In step S1, the basic information includes, but is not limited to, user name, age, gender, family medical history, and lifestyle, including but not limited to diet, exercise, and sleep.

3. The health check-up management platform according to claim 1, characterized in that: In step S1, the question-and-answer mechanism asks the user questions based on the characteristics of the inferred potential disease, and further infers the probability of the disease occurring based on the user's answer.

4. A health check-up management platform according to claim 1, characterized in that: In step S1, AR action guidance is carried out in conjunction with a question-and-answer mechanism. AR action guidance uses AR overlay guidance layers, such as red outlines, to mark the shooting range, and combines edge computing to detect image quality in real time, such as clarity and correctness of posture, in order to obtain the physical signs and symptoms of the user based on the inferred symptoms.

5. A health checkup management platform according to claim 1, characterized in that: In step S1, the calculation process for the probability of potential disease occurrence is as follows: Collect basic information: Obtain demographic characteristics such as age, gender, height, weight, family medical history such as diabetes and cancer, wearable device data such as heart rate and sleep duration, and lifestyle such as diet, exercise, and sleep through the user-end medical and health management APP; Retrieve historical physical examination data: After user authorization, retrieve cross-institutional physical examination reports through the blockchain interface, and analyze structured indicators such as blood pressure, blood sugar, blood lipids, and imaging conclusions such as "thyroid nodules TI-RADS 3". Question and Answer Data: Based on preliminary risk assessment, a progressive chain of questions is generated, such as regarding diabetes risk: "Have you experienced excessive thirst and urination in the past six months?" and "How often do symptoms occur?" Vital signs data: Through AR medical health management, users are guided to complete specific actions such as taking pictures of the thyroid area and recording videos of hand tremors, and vital signs such as nodule size and tremor frequency are extracted; Missing value handling: Key indicators such as fasting blood glucose are prompted to be supplemented by the user, while non-key indicators such as exercise frequency are filled with the median. Standardization processing: Numerical data such as BMI are normalized to the 0-1 range using Min-Max, and categorical data such as "family history" are converted to binary labels; Outlier correction: Identify extreme values ​​such as systolic blood pressure >200 mmHg using the IQR method, prompt the user for confirmation or replace it with the average value for the same age group; Extract common features: Calculate derived indicators, such as BMI = weight ÷ height², smoking index = number of cigarettes smoked per day × number of years, and exercise compliance rate = actual exercise time ÷ recommended exercise time; Extract disease-specific features: Screen key factors by disease type, such as diabetes associated with "fasting blood glucose, waist circumference, and frequency of polyuria"; thyroid disease associated with "gender and proportion of thyroid enlargement in AR images"; Extract interactive fusion features: Combine question and answer data with vital signs such as "frequency of nighttime acid reflux + degree of incoordination of throat muscles during swallowing" to generate a reflux risk index; Initial screening using a rule-based engine: Thresholds are set based on clinical guidelines, such as "fasting blood glucose ≥7.0 mmol / L + polydipsia and polyuria" → initial risk of diabetes is set at 80%; Multimodal model precision calculation: Input features into the Transformer fusion model and output probability values; Real-time updates: After users add data such as new blood glucose values ​​or complete AR actions, the above process is repeated immediately to update the risk probability.

6. A health check-up management platform according to claim 5, characterized in that: Disease risk grading: Based on probability values, it is divided into low risk <30%, medium risk 30%-70%, and high risk >70%. If the probability of a disease falls within the medium risk range, it is recommended that the user undergo a physical examination.

7. A health check-up management platform according to claim 1, characterized in that: The health checkup management platform also includes the following steps: S3. Post-examination stage: Integrate physical examination reports, genetic testing data, and wearable device information to generate a full life cycle health record, and push personalized health advice based on a knowledge graph-based health risk assessment model.

8. A health check-up management platform according to claim 7, characterized in that: In step S3, personalized health advice includes, but is not limited to, dietary adjustments and exercise plans.

Images