CN-121983209-A - Auxiliary analysis method for measurement data for health monitor based on AI large model

CN121983209ACN 121983209 ACN121983209 ACN 121983209ACN-121983209-A

Abstract

The invention relates to the technical field of artificial intelligence, and discloses an AI large model-based auxiliary analysis method for measurement data for a health monitor, which comprises the steps of obtaining physiological parameter measurement data of a user through the health monitor; the method comprises the steps of combining the measurement data, basic information of a user and historical health data into analysis context, deeply reading the analysis context by utilizing a pre-trained AI large model, wherein the AI large model can identify potential modes, relevance and risk signals in the data based on massive medical knowledge training, and finally, generating a comprehensive analysis report comprising health condition assessment, trend reading, abnormal early warning and personalized health advice. The intelligent health monitoring system overcomes the limitation that the traditional health monitor can only provide isolated data, provides intelligent and scene health insight comparable to a professional consultant for a user through strong reasoning and cognition capability of an AI large model, and greatly improves the value and practicability of home health monitoring.

Inventors

YIN GUANGQIANG
WEN PENG
YANG SHUNKUN
WANG XIANDE
GAO SANHONG

Assignees

喀什地区电子信息产业技术研究院

Dates

Publication Date: 20260505
Application Date: 20260112

Claims (10)

1. The auxiliary analysis method for the measurement data of the health monitor based on the AI large model is characterized by comprising the following steps: s1, synchronously acquiring multi-mode physiological measurement data of a user at a specific time point through a non-contact sensor and a contact sensor of a health monitor to construct a physiological parameter measurement data set of the current period of the user; s2, interactively fusing numerical feature vectors, image feature vectors and text feature vectors from the physiological parameter measurement data set to obtain an analysis context vector sequence; S3, extracting a historical time sequence data sequence corresponding to a current detection item from a historical health record, inputting the analysis context vector sequence and the historical time sequence data sequence into a pre-trained special AI large model together, and outputting a group of depth interpretation signals, wherein the depth interpretation signals comprise a current physiological state evaluation tag, an abnormal mode description, and key potential association and weak risk signal clusters identified from data; S4, analyzing sub-health conversion risks of the user based on the correlation of the key potential and the weak risk signal cluster and the change trend of the corresponding signals in the historical time sequence data sequence to obtain a sub-health risk index and a corresponding risk grade label; s5, based on the current physiological state evaluation label, the abnormal pattern description, the sub-health risk index and the risk grade label, automatically generating a structured personalized comprehensive health report according to a preset report template.
2. The AI-large-model-based auxiliary analysis method for measurement data for a health monitor according to claim 1, wherein the step of synchronously acquiring the multi-modal physiological measurement data of the user at a specific time point by the non-contact sensor and the contact sensor of the health monitor to construct a physiological parameter measurement data set of the current period of the user comprises the steps of: Based on an integrated contact sensor on a health monitor, acquiring a physiological parameter value set of a user in monitoring time, wherein the physiological parameter value set comprises heart rate, blood pressure, blood oxygen saturation and body surface temperature; based on a non-contact sensor integrated on the health monitor, acquiring user sign image data and environment perception data of a user in synchronization with the physiological parameter numerical data in the monitoring time; Integrating the physiological parameter value set aligned with the time stamp, the user sign image data and the environment perception data to obtain a multi-mode measurement data time sequence; and carrying out format standardization and structure reorganization on the aligned multi-mode measurement data to construct a physiological parameter measurement data set of the current period of the user, wherein the physiological parameter measurement data set comprises a numerical field, an image field and a description field.
3. The AI-large-model-based measurement data aided analysis method of claim 1, wherein the interactively fusing the numerical feature vectors, the image feature vectors, and the text feature vectors from the physiological parameter measurement dataset to obtain an analysis context vector sequence, comprising: extracting the numerical characteristics of the numerical field in the physiological parameter measurement data set to obtain a numerical characteristic vector; extracting the hierarchical features of the image type field in the physiological parameter measurement data set to obtain an image feature vector; natural language understanding and coding are carried out on the descriptive fields in the physiological parameter measurement data set, and text feature vectors are obtained; And carrying out feature dimension pair Ji Yingshe on the numerical feature vector, the image feature vector and the text feature vector, and carrying out cross-modal information weighted fusion on the numerical feature vector, the image feature vector and the text feature vector after dimension alignment based on an interactive attention fusion algorithm so as to obtain an analysis context vector sequence.
4. The AI-large-model-based measurement data aided analysis method for a health monitor of claim 3, wherein the extracting the numerical feature of the numerical field in the physiological parameter measurement data set to obtain the numerical feature vector includes: carrying out abnormal value cleaning and dimension standardization processing on numerical value fields in the physiological parameter measurement data set containing heart rate, blood pressure, blood oxygen saturation and body surface temperature so as to obtain a standardized numerical value sequence; and calculating time domain statistical characteristics and morphological characteristics of the standardized numerical sequence to obtain a numerical characteristic vector.
5. The AI-large-model-based measurement data aided analysis method of claim 3, wherein the extracting the hierarchical features of the image type field in the physiological parameter measurement data set to obtain the image feature vector includes: Preprocessing the image and video data contained in the image type field to obtain standardized image data; Carrying out hierarchical visual feature extraction on the standardized image data to obtain an image feature map for representing multi-scale semantic information of the image; And carrying out space dimension compression and vectorization conversion on the image feature map to obtain the image feature vector.
6. The AI-large-model-based measurement data aided analysis method for a health monitor of claim 3, wherein said natural language understanding and encoding of descriptive fields in said physiological parameter measurement dataset to obtain text feature vectors includes: Performing text normalization processing on the original text data contained in the descriptive field in the physiological parameter measurement data set to obtain a standardized vocabulary sequence; Based on a pre-constructed health field feature dictionary, feature matching and statistics counting are carried out on the standardized vocabulary sequence to obtain vocabulary level sparse feature vectors, and weight adjustment and dimension compression are carried out on the vocabulary level sparse feature vectors to obtain the text feature vectors.
7. The AI-large-model-based measurement data aided analysis method of claim 1, wherein extracting a historical time series data sequence corresponding to a current detection item from a historical health record, inputting the analysis context vector sequence and the historical time series data sequence together into a pre-trained special AI large model, and outputting a set of depth interpretation signals, comprises: based on the current detection item, regularizing the historical health record of the user to generate a standard historical time sequence data sequence isomorphic with the current physiological parameter measurement data set; based on a space-time fusion algorithm, fusing the analysis context vector sequence representing the current moment multi-mode state with the standard historical time sequence data sequence to construct a space-time enhanced model input tensor; And inputting the space-time enhanced model input tensor into a pre-trained special AI large model, and reasoning with a decoding layer through a multi-head attention mechanism in the model input tensor so as to output a depth interpretation signal group comprising a current physiological state evaluation label, an abnormal mode description, a key potential association and a weak risk signal cluster.
8. The AI-large-model-based measurement data aided analysis method for a health monitor of claim 7, wherein the analyzing the sub-health conversion risk of the user based on the key potential correlation with the weak risk signal cluster and the trend of the change of the corresponding signal in the historical time series data sequence to obtain a sub-health risk index and a corresponding risk class label comprises: Carrying out trend quantization on each signal element in the key potential association and the weak risk signal cluster to obtain a group of quantized risk index sets comprising signal intensity values, associated disease weight values and time sequence gradient values; performing risk quantitative evaluation on the quantitative risk index value to obtain a sub-health risk index for representing the possibility of conversion of the user to a specific disease and sub-health state; And carrying out multi-level threshold judgment and semantic mapping based on the numerical interval of the sub-health risk index and the composition property of the weak risk signal cluster so as to generate a corresponding risk level label.
9. The AI-large-model-based measurement data aided analysis method for a health monitor of claim 8, wherein said trend quantifying each signal element in said key potential associations and said weak risk signal clusters to obtain a set of quantified risk indicators including signal intensity values, associated disease weight values, and time-gradient values, comprises: Determining a signal strength value based on the magnitude of deviation of the physiological parameter measurement corresponding to each signal in the weak risk signal cluster from a personal baseline of the key potential association; Mapping to obtain an associated disease weight value according to the signal intensity value and the type of the signal element based on a pre-constructed medical knowledge base; Analyzing the occurrence frequency, duration time and numerical change rate of related physiological parameters of each signal corresponding to an abnormal mode in the weak risk signal cluster based on the historical time sequence data sequence, and calculating to obtain a time sequence gradient value; and combining the signal intensity value, the associated disease weight value and the time sequence gradient value to obtain a quantized risk index set.
10. The AI-large-model-based auxiliary analysis method of measurement data for health monitors according to claim 1, wherein the automatically generating a structured personalized comprehensive health report based on the current physiological state evaluation tag, the abnormal model description, the sub-health risk index and the risk class tag according to a preset report template comprises: Carrying out data alignment and structuring serialization processing on the current physiological state evaluation label, the abnormal pattern description, the sub-health risk index and the risk grade label so as to generate a standardized report input data packet; performing template matching, content filling and logic decision on the standardized report input data packet based on a preset report template to generate an original report content set comprising key interpretation, risk assessment and personalized advice; performing format rendering, visual element insertion and metadata attachment operations on the original report content set to generate and output the structured personalized comprehensive health report.

Description

Auxiliary analysis method for measurement data for health monitor based on AI large model Technical Field The invention relates to the technical field of artificial intelligence, in particular to an AI large model-based auxiliary analysis method for measurement data for a health monitor. Background Although the current health monitor can collect basic physiological data such as heart rate and blood pressure, the defect of insufficient data fusion capability exists generally, the prior art is used for processing single types of data such as numerical values, images and texts independently, an effective multi-mode information integration mechanism is lacked, potential association among data with different dimensions cannot be mined, an analysis result is only remained on an isolated data presentation level, comprehensive user health portraits are difficult to form, the single-mode analysis mode ignores the cooperative indication significance of physiological indexes, physical sign images, environment descriptions and other information, deep data support cannot be provided for health assessment, and the intelligent level of the monitor is limited. Meanwhile, the traditional health monitoring technology focuses on obvious abnormal alarms which occur, dynamic quantitative evaluation capability on sub-health conversion risks is lacking, the existing scheme often depends on fixed threshold judgment, early risk signals with weak strength and time duration cannot be captured, risk evolution trend cannot be analyzed by combining with user historical health tracks, early warning hysteresis of sub-health states is caused, users cannot take intervention measures in advance, in addition, the existing report generation mechanism lacks personalized adaptation capability, complex data cannot be converted into executable health guidance, requirements of users on accurate and scene health management are difficult to meet, and therefore how to improve personalized adaptation capability of the report generation mechanism, requirements of users on accurate and scene health management are met, and the problem to be solved urgently is solved. Disclosure of Invention The invention provides an AI large model-based auxiliary analysis method for measurement data for a health monitor, which aims to solve the problems in the background technology. In order to achieve the above object, the method for assisting in analyzing measurement data for health monitor based on AI large model provided by the invention comprises the following steps: s1, synchronously acquiring multi-mode physiological measurement data of a user at a specific time point through a non-contact sensor and a contact sensor of a health monitor to construct a physiological parameter measurement data set of the current period of the user; s2, interactively fusing numerical feature vectors, image feature vectors and text feature vectors from the physiological parameter measurement data set to obtain an analysis context vector sequence; S3, extracting a historical time sequence data sequence corresponding to a current detection item from a historical health record, inputting the analysis context vector sequence and the historical time sequence data sequence into a pre-trained special AI large model together, and outputting a group of depth interpretation signals, wherein the depth interpretation signals comprise a current physiological state evaluation tag, an abnormal mode description, and key potential association and weak risk signal clusters identified from data; S4, analyzing sub-health conversion risks of the user based on the correlation of the key potential and the weak risk signal cluster and the change trend of the corresponding signals in the historical time sequence data sequence to obtain a sub-health risk index and a corresponding risk grade label; s5, based on the current physiological state evaluation label, the abnormal pattern description, the sub-health risk index and the risk grade label, automatically generating a structured personalized comprehensive health report according to a preset report template. In a preferred embodiment, the method for synchronously acquiring the multi-mode physiological measurement data of the user at a specific time point through the non-contact and contact sensors of the health monitor to construct a physiological parameter measurement data set of the current period of the user comprises the following steps: Based on an integrated contact sensor on a health monitor, acquiring a physiological parameter value set of a user in monitoring time, wherein the physiological parameter value set comprises heart rate, blood pressure, blood oxygen saturation and body surface temperature; based on a non-contact sensor integrated on the health monitor, acquiring user sign image data and environment perception data of a user in synchronization with the physiological parameter numerical data in the monitoring time; Integrating the physiologic