Disease detection intelligent risk assessment method and system

Abstract

The invention discloses an intelligent risk assessment method and system for disease detection, and the method comprises the steps: delimiting a normal fluctuation interval based on the steady state of an individual through collecting the historical physiological measurement values of the individual, extracting daily activity characteristics through combining behavior data, and distributing an initial risk weight according to a preset proportion; and continuously comparing the current physiological index with an individual normal fluctuation interval by accessing a real-time data stream, and screening unexplained abnormal signals in combination with environmental parameters. The method gets rid of dependence on a general standard and single measurement, and can identify an abnormal trend deviating from a self steady state when a physiological index does not break through a general medical threshold value by utilizing individualized baseline and multi-source data fusion. By establishing a dynamic reference system and a context screening mechanism, the system can start a risk assessment process at the initial stage of an abnormal trend, so that the timeliness and pertinence of early warning are remarkably improved, and early capture of potential health risks is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of health risk assessment, and particularly relates to a disease detection intelligent risk assessment method and system. BACKGROUND

[0002] In the existing health risk assessment practice, the assessment method based on the physical examination data at a fixed time point and the general medical standard is generally adopted. After an individual completes physical examination at a medical institution, the physiological indicators such as heart rate, blood pressure and blood sugar are compared with the unified reference interval one by one for judgment, and the individual is marked as abnormal if the indicators exceed the range. The process is usually carried out once a year, and the assessment result depends on the comparison between the single measurement value and the static threshold value, and does not involve the tracking of the long-term trend of the individual. In addition, the behavior data or environmental exposure information from the wearable device is rarely integrated in the assessment process, and a dynamic analysis path for the correlation between the lifestyle and the physiological state is not established. Due to the low data collection frequency, the non-personalized reference standard and the single information dimension, the assessment result cannot reflect the short-term fluctuations or progressive changes of the individual in daily life.

[0003] This method has a significant technical problem: it is difficult to identify the early abnormal trend of the individual's health status in time, the assessment result lags behind the actual physiological change, and the pre-warning for the sudden or progressive health risk cannot be realized. For example, when the fasting blood sugar value of the individual continuously and slowly rises in several months, but is always near the upper limit of the general reference interval, the traditional method does not give a warning because the threshold value is not broken, and the individual is not discovered until obvious over-standard or clinical symptoms appear, which misses the best intervention opportunity. SUMMARY

[0004] The purpose of the present application is to provide a disease detection intelligent risk assessment method and system, which establishes a dynamic reference system and a context screening mechanism, so that the system can start the risk assessment process at the initial stage of the abnormal trend, significantly improves the timeliness and pertinence of the warning, and realizes the early capture of the potential health risk, to solve the problems proposed in the background.

[0005] To achieve the above purpose, the present application adopts the following technical scheme: a disease detection intelligent risk assessment method, comprising the following steps: Collecting the birth month, gender, family history of genetic diseases and records of previous major diseases of the individual, and generating a unique identity; Calling the physiological measurement values in the past 6 months in the healthy state, calculating the mean and standard deviation of each index, and setting the individual normal fluctuation interval as the mean ± 1.5 times the standard deviation; Integrating the behavior data for 30 consecutive days, analyzing the time distribution of the daily walking number, sleep duration and dietary regularity, and extracting the daily activity characteristics; Based on the information corresponding to the identity identifier, the normal fluctuation range and the characteristics of daily activities, the initial risk weights for different disease categories are allocated according to a preset ratio; By connecting to real-time data streams from wearable devices and environmental sensors, current physiological indicators are compared with the individual's normal fluctuation range, deviations are marked, and unexplained abnormal signals are filtered in conjunction with environmental parameters.

[0006] Preferably, the extraction of daily activity features includes: Calculate the mean and standard deviation for each behavioral indicator; The ratio of standard deviation to mean is used to measure the stability of behavioral rhythms; Identify behavioral patterns that deviate from the mean by more than 20% over seven consecutive days; Behaviors with low stability and persistent deviations are labeled as rhythmic perturbations.

[0007] Preferably, the allocation of initial risk weights for different disease categories according to a preset ratio includes: Set the disease predisposition corresponding to a family history of genetic diseases as the first vector; The relative distance between each physiological indicator and the midpoint of its normal fluctuation range is set as the second vector; Transform the stability of behavioral rhythms into a third vector; The initial risk weight distribution is obtained by weighted summation of the three vectors.

[0008] Preferably, the step of marking deviation items and filtering unexplained anomalous signals in conjunction with environmental parameters includes: The physiological indicators are compared with the individual's normal fluctuation range every 5 minutes; When the current value exceeds the individual's normal fluctuation range, it is marked as an initial deviation; Simultaneously acquire ambient temperature, humidity, and air quality data. If the environmental parameters are within a known interference range, reduce the risk level of the deviation.

[0009] Preferably, the step of marking deviation items and filtering unexplained anomalous signals in conjunction with environmental parameters further includes: For the unexplained anomalous signals that were screened out, time-series segments were formed by tracing back the measurement records of the past 7 days. Apply a 24-hour sliding window to the time series segment, calculate the difference between the final value and the initial value within the window, divide by the time span, and obtain the local rate of change. The trend curve is obtained by least-squares fitting of the time series segments using a quadratic polynomial; The predicted value for the next 12 hours is extrapolated from the trend curve and compared with the upper limit of the individual's normal fluctuation range to determine whether there is a trend risk.

[0010] Preferably, the step of using a quadratic polynomial to perform least-squares fitting on the time series segments to obtain the trend curve includes: The quadratic polynomial is set to be a linear combination of the time square term, the time linear term, and the constant term. The multipliers of the three terms are adjusted to minimize the sum of squared errors between the fitted result and the actual measured value. The term with the largest multiplier is retained as the dominant trend feature, and the dominant trend feature is used to determine whether the change in the indicator is an acceleration or a slowdown.

[0011] Preferably, the step of extrapolating the predicted value for the next 12 hours based on the trend curve and comparing it with the upper limit of the individual's normal fluctuation range to determine whether there is a trend risk includes: Extract dietary records, exercise duration, sleep quality, and indoor / outdoor temperature data from the same period. Calculate the ratio of the covariance to the standard deviation of each data point and abnormal physiological indicator over 7 days; Data with ratios greater than 0.5 are marked as significantly relevant variables; The presence or absence of significantly relevant variables determines whether the risk originates from lifestyle or underlying pathological processes.

[0012] Preferably, the step of determining whether a risk originates from lifestyle or an underlying pathological process based on the presence or absence of a significantly relevant variable includes: If the attribution result is lifestyle influence, then the current disease category weight remains unchanged; If the attribution result is an underlying pathological process, the weight of the corresponding disease category will be increased by 10% to 15%. Other category weights are compressed proportionally to maintain a total of 100%; Three-level early warning thresholds are set based on the updated weights and trend risk scores.

[0013] Preferably, the step of setting a three-level early warning threshold based on the updated weights and trend risk scores includes: Health advice is sent when the trend risk score is greater than 0.5; When the trend risk score is greater than 0.8, a medical examination is recommended. When the trend risk score is greater than 1.2, notify the emergency contact person. Record the warning content, trigger time, and supporting data, and enter the pending feedback state.

[0014] On the other hand, this invention proposes an intelligent risk assessment system for disease detection, comprising: The data acquisition unit is used to obtain an individual's date of birth, gender, family history of genetic diseases, records of past major illnesses, and physiological measurements of their health status over the past 6 months. The interval generation unit is used to calculate the mean and standard deviation based on physiological measurements, and to define the normal fluctuation range of an individual with the mean ± 1.5 times the standard deviation. The feature extraction unit is used to integrate behavioral data from 30 consecutive days, analyze the temporal distribution of average daily steps, sleep duration and dietary regularity, and extract features of daily activities. The weighting allocation unit is used to combine the information corresponding to the identity identifier, the individual's normal fluctuation range and daily activity characteristics, and allocate the initial risk weights of different disease categories according to a preset ratio. The anomaly screening unit is used to access the real-time data streams from wearable devices and environmental sensors, compare the current physiological indicators with the individual's normal fluctuation range, mark deviations, and filter out unexplained abnormal signals in conjunction with environmental parameters.

[0015] Technical effects and advantages of the present invention: The intelligent risk assessment method and system for disease detection proposed in this invention have the following advantages compared with the prior art: This invention delineates a normal fluctuation range based on an individual's homeostasis by collecting historical physiological measurements. It then extracts daily activity characteristics from behavioral data and assigns initial risk weights according to a preset ratio. By continuously comparing current physiological indicators with the individual's normal fluctuation range through access to real-time data streams, and combining environmental parameters, it filters out unexplained abnormal signals. This approach eliminates reliance on universal standards and single measurements. By utilizing individualized baselines and multi-source data fusion, it can identify abnormal trends deviating from homeostasis before physiological indicators exceed general medical thresholds. By establishing a dynamic reference system and contextual filtering mechanism, the system can initiate a risk assessment process at the initial stage of abnormal trends, significantly improving the timeliness and specificity of early warnings and enabling early detection of potential health risks. Attached Figure Description

[0016] Figure 1 This is a flowchart of an intelligent risk assessment method for disease detection according to the present invention; Figure 2 This is a block diagram of an intelligent risk assessment system for disease detection according to the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] This invention provides, for example Figure 1This paper presents an intelligent risk assessment method for disease detection. By continuously collecting an individual's physiological parameters, environmental exposure information, and daily behavioral characteristics, it constructs a dynamic assessment path that automatically adjusts assessment weights and thresholds over time. This path can identify early signals of potential health risks and trigger warning responses at key turning points. The entire process is data-driven, emphasizing the continuity of information flow and the progressive nature of state progression to ensure that the assessment conclusions remain synchronized with the individual's current state. The method includes the following steps: Collect an individual's date of birth, gender, family history of genetic diseases, and records of past major illnesses to generate a unique identity identifier; establish an individual-specific data link anchor to achieve orderly collection and long-term tracking of multi-source health information, avoid data confusion, and provide identity protection for subsequent personalized analysis.

[0019] It retrieves physiological measurements taken during the past 6 months when the individual was in a healthy state, calculates the mean and standard deviation of each indicator, and defines the normal fluctuation range of an individual with the mean ± 1.5 times the standard deviation. This reflects the physiological homeostasis range of an individual in real life, reduces the probability of misjudgment caused by physiological fluctuations, and improves the individual suitability of abnormality detection.

[0020] This study integrates behavioral data from 30 consecutive days, analyzing the temporal distribution of average daily steps, sleep duration, and dietary regularity to extract daily activity characteristics. Specifically, it calculates the mean and standard deviation for each behavioral indicator; uses the ratio of standard deviation to mean to measure the stability of behavioral rhythms; identifies behavioral patterns that deviate from the mean by more than 20% over seven consecutive days; and marks behaviors with low stability and persistent deviations as rhythmic perturbations. This quantifies the inherent fluctuation patterns of an individual's life rhythm, identifies atypical behavioral changes, provides an explanatory basis for physiological state fluctuations, and enhances the environmental relevance of risk assessment.

[0021] By combining information corresponding to identity markers, normal fluctuation ranges, and daily activity characteristics, initial risk weights for different disease categories are allocated according to a preset ratio. Specifically, this includes: setting the disease predisposition corresponding to family history of genetic diseases as the first vector; setting the relative distance between various physiological indicators and the midpoint of their normal fluctuation ranges as the second vector; converting the stability of behavioral rhythms into a third vector; and then weighting and summing the three vectors to obtain the initial risk weight distribution. By integrating three heterogeneous types of information—genetic background, physiological baseline, and behavioral patterns—an initial risk distribution that reflects individual differences is generated, making the assessment starting point closer to the true health profile and enhancing the directionality of subsequent risk extrapolation.

[0022] The system integrates real-time data streams from wearable devices and environmental sensors, comparing current physiological indicators with the individual's normal fluctuation range. Deviations are flagged, and unexplained abnormal signals are filtered using environmental parameters. Specifically, this involves: comparing physiological indicators with the individual's normal fluctuation range every 5 minutes; marking a preliminary deviation when the current value exceeds the individual's normal fluctuation range; simultaneously acquiring environmental temperature, humidity, and air quality data; reducing the risk level of deviations if environmental parameters are within known interference ranges; for the filtered unexplained abnormal signals, backtracking measurement records from the past 7 days to form a time series segment; applying a 24-hour sliding window to the time series segment, calculating the difference between the final and initial values ​​within the window divided by the time span to obtain the local rate of change; performing least-squares fitting on the time series segment using a quadratic polynomial to obtain a trend curve; extrapolating the predicted values ​​for the next 12 hours based on the trend curve and comparing them with the upper limit of the individual's normal fluctuation range to determine if there is a trend risk.

[0023] It enables continuous dynamic monitoring of physiological states, distinguishes between environmentally induced physiological fluctuations and potential pathological signals, focuses on abnormal changes not explained by external factors, identifies progressive risks through trend extrapolation, and improves the foresight and accuracy of early warning.

[0024] This method employs a quadratic polynomial to perform least-squares fitting on time segments to obtain trend curves. This involves: defining the quadratic polynomial as a linear combination of the output variable and the time square term, the time linear term, and a constant term; adjusting the multipliers of these three terms to minimize the sum of squared errors between the fitted result and the actual measured value; retaining the term with the largest multiplier as the dominant trend feature; and using this dominant trend feature to determine whether the indicator change is accelerating or slowing down. This method captures the non-linear characteristics of physiological indicator changes, identifies trends of accelerated deterioration or mitigation, improves sensitivity to pre-disease states, and avoids risky misjudgments caused by linear assumptions.

[0025] The process involves extrapolating predicted values ​​for the next 12 hours based on trend curves and comparing them with the upper limit of an individual's normal fluctuation range to determine the existence of trend risks. This includes: extracting dietary records, exercise duration, sleep quality, and indoor / outdoor temperature data from the same period; calculating the ratio of the covariance to the standard deviation of each data point and abnormal physiological indicator over 7 days; marking data with a ratio greater than 0.5 as significantly correlated variables; and determining whether the risk originates from lifestyle or a potential pathological process based on the presence or absence of significantly correlated variables. Specifically, if the attribution result is lifestyle influence, the current disease category weight remains unchanged; if the attribution result is a potential pathological process, the weight of the corresponding disease category is increased by 10% to 15%; the weights of other categories are proportionally compressed to maintain a total of 100%; and a three-level warning threshold is set based on the updated weights and trend risk score. By differentiating the causes of risks, focusing more on trend deviations caused by non-lifestyle factors, and dynamically adjusting the assessment focus, the warning response is matched with the severity of potential health threats, thereby improving the rationality of resource allocation.

[0026] For example, a three-tiered early warning threshold is set based on the updated weights and trend risk scores, including: pushing health advice when the trend risk score is greater than 0.5; reminding patients to seek medical attention when the trend risk score is greater than 0.8; and notifying emergency contacts when the trend risk score is greater than 1.2. The warning content, trigger time, and supporting data are recorded, and the system enters a pending feedback state. This establishes a tiered response mechanism that matches the severity of the risk, enabling precise distribution of early warning information, ensuring that intervention measures are synchronized with changes in individual conditions, and improving the timeliness and operability of health management.

[0027] On the other hand, this invention proposes an intelligent risk assessment system for disease detection, such as... Figure 2 As shown, it includes: The data acquisition unit is used to obtain an individual's date of birth, gender, family history of genetic diseases, records of past major illnesses, and physiological measurements of their health status over the past 6 months. The interval generation unit is used to calculate the mean and standard deviation based on physiological measurements, and to define the normal fluctuation range of an individual with the mean ± 1.5 times the standard deviation. The feature extraction unit is used to integrate behavioral data from 30 consecutive days, analyze the temporal distribution of average daily steps, sleep duration and dietary regularity, and extract features of daily activities. The weighting allocation unit is used to combine the information corresponding to the identity identifier, the individual's normal fluctuation range and daily activity characteristics, and allocate the initial risk weights of different disease categories according to a preset ratio. The anomaly screening unit is used to access the real-time data streams from wearable devices and environmental sensors, compare the current physiological indicators with the individual's normal fluctuation range, mark deviations, and filter out unexplained abnormal signals in conjunction with environmental parameters.

[0028] In addition, the aforementioned units are also used to implement other steps of the intelligent risk assessment method for disease detection as described above, as follows: Step 1: Establish an individual's basic health profile and initialize assessment benchmarks This step aims to construct an initial health description framework for each assessment subject, serving as a benchmark for subsequent dynamic assessments. This framework not only includes static demographic information but also incorporates the typical physiological fluctuations and behavioral preference patterns of an individual in a healthy state. By setting initial reference intervals, a comparative basis is provided for subsequent bias identification.

[0029] Step 1.1: Collect individual static attribute information and construct identity identifiers: At the outset of the assessment process, basic individual attribute information is first acquired, including date of birth, gender, family history of genetic diseases, and records of past major illnesses. This information is obtained through an authorized electronic health record interface or user-inputted data and encrypted to generate a unique identification code. This code is used throughout the entire assessment process to link all subsequent dynamic data streams, ensuring the uniqueness and traceability of information ownership. The data collection process follows the principle of minimum necessity, acquiring only core fields related to health risks and avoiding redundant storage of irrelevant information. This identification code is not only used for data aggregation but also serves as the starting point for subsequent personalized threshold settings, such as presetting a baseline probability range for cardiovascular events based on gender and age. The generation of this identification code uses hash encryption to ensure privacy and security while supporting cross-device data synchronization.

[0030] Step 1.2: Obtain individual baseline physiological indicators and determine the normal fluctuation range: After identity verification is established, the system retrieves physiological data recorded by the individual in medical institutions or home devices over the past six months, filters out measurements within a clinically recognized "healthy state," and forms a baseline dataset. The included indicators include, but are not limited to, resting heart rate, systolic blood pressure, fasting blood glucose, blood oxygen saturation, and body temperature. For each indicator, its mean and standard deviation are calculated, and the normal fluctuation range for the individual is defined as mean ± 1.5 times the standard deviation. For example, if a user's average resting heart rate over the past six months is 72 beats / minute and the standard deviation is 6, then their normal range is set to 63 to 81 beats / minute. This range differs from general medical standards; instead, it reflects the individual's own homeostatic range, improving the sensitivity of subsequent anomaly detection. This range will serve as the initial reference for subsequent real-time data comparison.

[0031] To achieve a mathematical expression for this interval, the following formula is introduced: ; in, Indicates the first The normal fluctuation range of a physiological indicator For its historical average, The standard deviation is used as the reference value. The principle behind this formula is to utilize the central tendency and dispersion of an individual's historical data to construct a boundary that encompasses their natural physiological fluctuations. Compared to fixed medical thresholds (such as "normal heart rate 60–100"), this range better reflects individual differences and reduces misjudgments caused by physiological fluctuations. For example, an athlete's resting heart rate consistently remains at 55 beats per minute, which, although below the general lower limit, is still included in their personal normal range due to the stability of their historical data, avoiding unnecessary alarms.

[0032] Step 1.3: Analyze historical behavioral patterns and extract features of daily activities: Building upon the established physiological baseline, historical behavioral data from wearable devices or mobile applications is further integrated, encompassing daily step count, sleep duration, nighttime turning frequency, dietary regularity, and screen time. Through time-series analysis of 30 consecutive days of data, typical individual behavioral rhythms are identified, such as bedtime concentrated between 10:30 PM and 11:15 PM, and morning activity starting between 6:45 AM and 7:15 AM. These behavioral characteristics are transformed into a set of time distribution parameters to characterize the stability of an individual's circadian rhythm. If a behavioral indicator deviates from its historical mean by more than 20% over seven consecutive days, it is labeled as a "circadian rhythm disturbance." This feature set constitutes a quantitative expression of an individual's lifestyle, providing a basis for subsequent inferences about the interaction between environment and behavior.

[0033] To quantify the stability of behavioral rhythms, the coefficient of variation (CV) is introduced as a metric: ; in, Indicates the first The coefficient of variation of the behavioral indicators Its standard deviation, Its average. The principle behind this formula is to evaluate the consistency of behavior through relative dispersion. If a user's average daily steps are... This indicates that its activity level is highly stable; if This indicates significant fluctuations in behavioral patterns, potentially influenced by work rhythm or psychological state. This parameter will be used for subsequent dynamic adjustment of risk weights; for example, highly variable users are more prone to physiological stress responses during sudden environmental changes.

[0034] Step 1.4: Set initial risk weights by combining static and dynamic baselines: Based on the aforementioned three types of information—identity attributes, physiological baseline, and behavioral characteristics—the system generates an initial risk weight distribution. This distribution is presented in percentage form, providing a preliminary allocation of potential risks for different disease categories. For example, if an individual has a family history of diabetes and their fasting blood glucose baseline is close to the upper limit, the weight for metabolic diseases is increased to 28%; if an individual is sedentary and has low heart rate variability, the weight for circulatory system-related diseases is set to 25%. The weight allocation process uses a weighted summation method, with the contribution of each dimension determined by preset empirical proportions: genetic factors account for 30%, physiological baseline for 40%, and behavioral characteristics for 30%.

[0035] To achieve the mathematical composition of weights, an initial weight vector is defined. ; in, This represents a genetic risk vector (such as disease predisposition corresponding to family history). This is the physiological baseline deviation vector (the relative distance between each indicator and the midpoint of its normal range). The behavioral instability vector (by...) (Transformed) This represents the contribution ratio of each dimension. The principle behind this formula is to unify multi-source heterogeneous information into a comparable numerical space, allowing the influence of different dimensions to be superimposed. For example, if a user... Blood sugar levels are slightly elevated. If diabetes has a high weight, then the combined weighting of the two significantly increases the overall weight of metabolic diseases. This initial weight vector This will serve as the starting point for subsequent dynamic updates, ensuring that the evaluation path begins from an individualized starting point.

[0036] Step 2: Collect multi-source health data in real time and perform preliminary anomaly screening. After the initial profile is built, the system enters the continuous monitoring phase. This step focuses on acquiring real-time physiological, behavioral, and environmental data of individuals from multiple terminal devices and performing preliminary screening to identify abnormal signals that significantly deviate from the baseline. This process ensures that only data with potential risk significance enters the in-depth analysis process, avoiding the unnecessary consumption of system resources.

[0037] Step 2.1: Connect the wearable device and environmental sensor data streams: Based on the identity established in Step 1, the system automatically binds user-authorized wearable devices (such as smartwatches and heart rate monitors), home medical devices (such as electronic blood pressure monitors and blood glucose meters), and environmental monitoring devices (such as indoor thermometers and hygrometers and air quality detectors). These devices upload data at minute or second-level frequencies, forming a continuous data stream. Each data packet includes a timestamp, device type identifier, and checksum to ensure the source is trustworthy. Data access uses a lightweight communication protocol to reduce device power consumption and supports a breakpoint resume mechanism to prevent data loss due to network fluctuations. This data stream constitutes the raw input for subsequent analysis, and its completeness directly affects the timeliness and coverage of the assessment.

[0038] Step 2.2: Baseline comparison and deviation labeling of real-time physiological indicators: Upon receiving real-time physiological data, the system immediately invokes the normal fluctuation range established in step one. Each indicator is compared. If the current value... fall out If the deviation is within a certain range, it is marked as "preliminary deviation," and the direction (too high or too low) and magnitude of the deviation are recorded. For example, if a user's real-time heart rate is 85 beats per minute, and their... If the heart rate is too high, it is determined to be "high," with a deviation of 4 beats per minute. This comparison process is performed every 5 minutes, forming a continuous deviation sequence. For brief fluctuations (such as an increase in heart rate after exercise), the system considers the context of the activity state to determine whether it is a physiological response, avoiding false labeling. This process relies on the definition in step 1.2. The interval ensures the individualization of the comparison standard.

[0039] Step 2.3: Determine external interference factors based on environmental parameters: After identifying deviations in physiological indicators, the system simultaneously retrieves environmental sensor data to analyze whether external interference factors exist. For example, if the heart rate is high while the ambient temperature exceeds 30°C or the PM2.5 concentration is higher than 75 μg / m³, the deviation is determined to be likely caused by high temperature or air pollution, rather than an intrinsic pathological change. To this end, the system establishes an environmental impact mapping table to record the typical influence directions of common environmental variables on physiological indicators. If the current environmental parameter is within a known interference range, the physiological deviation is marked as "environmentally relevant," reducing its risk level; if the environment is stable but the physiological indicator is abnormal, it is marked as "possibly endogenous," proceeding to the next level of analysis. This judgment relies on the environmental data stream accessed in step 2.1 to ensure contextual integrity.

[0040] Step 2.4: Generate a preliminary set of abnormal events and initiate a deep analysis request: After completing the combined physiological and environmental screening, the system summarizes all unexplained abnormal signals and generates a preliminary set of abnormal events. This set only includes physiological indicators that consistently deviate from baseline under stable conditions for more than 15 minutes. For example, a user whose resting heart rate remains above 83 beats per minute for 20 consecutive minutes under conditions of 24°C room temperature and good air quality is included. Each event includes metadata such as indicator name, deviation value, duration, and time period. This event set serves as a "start signal" to trigger in-depth analysis, notifying subsequent processes to prepare to receive data for processing. This process relies on the outputs of steps 2.2 and 2.3 to ensure that only signals with potential pathological significance are transmitted, avoiding information overload.

[0041] Step 3: Construct the time-series evolution path and identify trend changes This step goes beyond single-point anomaly detection and shifts to analyzing the long-term evolution trends of physiological indicators. By constructing time-series paths for various individual indicators, it identifies potential risk patterns of slow progression or periodic fluctuations, compensating for the lag in static assessments for progressive lesions.

[0042] Step 3.1: Extract historical time-series data fragments of abnormal indicators: Receive the preliminary set of abnormal events generated in step two Then, for each abnormal indicator, the system traces back the complete measurement records of the past 7 days to form a time series segment. ,in This is the current value. This segment covers at least 100 valid data points to ensure the statistical reliability of the trend analysis. Data points are sorted by time, and obviously erroneous values ​​(such as heart rate 0 or 300) are removed. This segment forms the input basis for trend analysis, and its time span is sufficient to capture short-term fluctuations and medium-term trends. This process directly references... The list of abnormal indicators ensures that the analysis is targeted at specific objects.

[0043] Step 3.2: Calculate the local rate of change using a sliding window: To quantify the rate of change of the indicators, the system performs time-series segment analysis. A sliding window with a width of 24 hours is applied, and the local rate of change is calculated window by window. The rate of change is defined as the difference between the final value and the initial value within the window divided by the time span. ; in, For the first The rate of change for each window is expressed in units per hour. This formula reflects the net trend of the indicator's change within a day. And if it continues to rise, it indicates that the indicator is showing an accelerating growth trend; if Drastic fluctuations indicate poor stability. For example, if fasting blood glucose levels... Three consecutive days of positive and increasing values ​​may indicate gradual decompensation of pancreatic islet function. This calculation relies on the data extracted in step 3.1. Fragments ensure the continuity of input data.

[0044] Step 3.3: Fit a nonlinear trend curve and predict short-term trends: After obtaining the rate of change sequence, the system uses a quadratic polynomial to process the original time series data. By performing least-squares fitting, the trend curve is obtained: ; Among them, coefficient Determined by the fitting process. This curve can capture acceleration or deceleration patterns. If This indicates that the indicator is showing an accelerating upward trend; if This indicates a slowdown or decline in growth. Based on this curve, extrapolate the forecast for the next 12 hours. and the normal range Upper limit comparison. If the predicted value exceeds the upper limit, it is judged as "trend risk". This method is superior to linear prediction and can identify nonlinear deterioration earlier. This process relies on the rate of change characteristics revealed in step 3.2. If the rate of change itself shows an upward trend, a nonlinear model is more likely to be used.

[0045] Step 3.4: Generate trend risk signals and update event priorities: Based on the trend analysis results, the system generates a trend risk score for each abnormal indicator. ; in, The empirical weights are 0.6 and 0.4, with the first term reflecting the acceleration intensity and the second term reflecting the degree of prediction deviation. The higher the value, the more urgent the risk. Based on this, the system analyzes the initial set of abnormal events. The entries are reordered, with high-scoring events given priority. This scoring mechanism ensures that rapidly progressing potential problems receive greater attention, optimizing resource allocation. This process directly references the fitted parameters and predicted values ​​from step 3.3, achieving a closed loop from trend identification to priority determination.

[0046] Step 4: Integrate multi-dimensional information for risk attribution analysis: After identifying trending risks, this step aims to explore their underlying causes. By integrating physiological, behavioral, and environmental data, a multidimensional correlation network is constructed to determine whether the risk is more likely to stem from lifestyle, environmental exposure, or underlying pathological processes.

[0047] Step 4.1: Extract the behavioral and environmental variables associated with risk indicators: For the high-priority risk events identified in step three, the system extracts behavioral and environmental data recorded concurrently to form a set of related variables. For example, if fasting blood glucose shows an upward trend, it extracts dietary records (carbohydrate intake), exercise duration, sleep quality, indoor and outdoor temperatures, etc., from the same period. The data is aligned to the same time granularity (e.g., hourly averages) to ensure comparability. This set of variables constitutes the input for attribution analysis, and its completeness directly affects the depth of the analysis. This process directly references the high-priority events identified in step 3.4.

[0048] Step 4.2: Calculate the time-series correlation between each variable and the risk indicator: For each associated variable Calculate its correlation with risk indicators Pearson correlation coefficient over the past 7 days ; in, For covariance, The standard deviation is denoted as . It is considered significantly correlated. Furthermore, if the increase in carbohydrate intake coincides with the increase in blood glucose, it suggests that dietary factors may dominate the trend; if the correlation is weak, the pathological hypothesis should be considered. This formula quantifies the strength of the linear association between variables, providing a statistical basis for attribution. This calculation relies on the aligned dataset from step 4.1.

[0049] Step 4.3: Classify the impact categories based on correlation strength: according to Size categorizes influencing sources into three types: dominant factors ( ), Synergistic factors Irrelevant factors If a risk event has no dominant or synergistic factor, it is classified as "of unknown cause," suggesting an underlying pathological process. For example, if blood glucose rises but diet, exercise, and sleep remain stable, it is more likely due to progression of insulin resistance. This classification mechanism guides subsequent interventions. This process directly references the correlation coefficient results from step 4.2.

[0050] Step 4.4: Generate attribution conclusions and update risk nature labels: Based on the comprehensive correlation analysis results, the system generates attribution conclusions, such as "mainly affected by irregular eating habits" or "cause unknown, suspected pathological progression." These conclusions are then appended to the risk events, forming an updated event set. The attribution results directly influence the direction of subsequent intervention recommendations. This process relies on the classification output of step 4.3 to ensure that the conclusions are based on evidence.

[0051] Step 5: Dynamically adjust evaluation weights and generate personalized early warnings Based on the attribution analysis results, this step dynamically adjusts the initial risk weights and generates tiered early warning information, enabling the self-evolution of the assessment system.

[0052] Step 5.1: Adjust the weights of relevant diseases based on the attribution results: Referring to the attribution conclusion in step 4.4, if a disease category is identified as high-risk (e.g., "increased metabolic risk"), then it will be included in the weight vector. The proportion will be increased by 10% to 15%. The adjustment formula is as follows: ; in, The original weights, To adjust the magnitude, the weights of other categories are proportionally compressed, maintaining a total of 100%. This update makes the assessment system more focused on active risks and improves the sensitivity of early warnings. This process directly relies on attribution conclusions.

[0053] Step 5.2: Set dynamic early warning thresholds and trigger them in stages: Based on updated weights and trend risk scores Set up a three-level early warning system: Attention Level: Push health advice; Warning level: This serves as a reminder to seek medical attention and undergo examination. Emergency Level: Notify the emergency contact person.

[0054] The threshold fluctuates dynamically with the weight; the threshold decreases for high-weight categories, making it easier to trigger alarms.

[0055] Step 5.3: Generate and push personalized alert messages: Based on the attribution conclusions and warning levels, natural language messages are generated, such as "Your blood sugar trend shows an increase, which may be related to your recent diet. We recommend adjusting your carbohydrate intake." These messages are pushed through channels such as the app and SMS to ensure timely delivery.

[0056] Step 5.4: Record the warning log and prepare for the next round of assessment: All early warning information is stored in an encrypted log, including the time, content, and supporting data. The system resets its analysis state, awaiting the next abnormal event to trigger, thus achieving closed-loop operation.

[0057] Step Six: Provide a feedback channel for the intervention and verify the effectiveness of the assessment. The assessment process does not end after the risk warning is sent. To ensure the entire process has the ability to self-verify and continuously optimize, this step introduces a two-way information exchange mechanism, allowing individuals to provide feedback on the warning content and record their subsequent behavioral adjustments and physiological changes. By mapping the intervention results back into the assessment logic, the system can verify the rationality of the initial judgment and provide experience support for decision-making in similar future situations. This mechanism breaks the traditional one-way output model, enabling the assessment system to learn and adapt.

[0058] Step 6.1: Receive user responses and behavior records to the warning information: After the alert message generated in step five is pushed out, the system opens a 72-hour feedback window. During this period, users can select preset response options through the mobile application interface, such as "Diet has been adjusted," "Exercise has been increased," "Medical examination has been conducted," or "I believe it's a false alarm." Simultaneously, the system automatically monitors the user's subsequent behavioral data stream, such as changes in steps, sleep duration, and updates to their diet log, as objective behavioral evidence. For example, if the alert suggests reducing carbohydrate intake, and the user's carbohydrate intake decreases by 30% over the next two days, it is considered a valid response. This process directly references the alert content pushed out in step five, ensuring the correspondence between feedback and specific recommendations. Response data and behavioral logs together constitute the intervention evidence set, providing input for subsequent effect evaluation.

[0059] Step 6.2: Track the trends of key physiological indicators after intervention: After the user completes the response, the system continuously tracks the physiological indicators related to the warning and observes their evolution over the next 48 to 72 hours. Specifically, it extracts the time from the intervention initiation point. to Hourly measurements form post-intervention time series segments. This segment is related to the preceding trend segment in step three. Compare and analyze whether the direction of change has reversed. For example, if blood sugar showed an upward trend before the warning ( ), and after intervention, the trend stabilized or declined ( If the intervention is effective, it can be preliminarily judged that the intervention is effective. This analysis relies on the intervention start time point determined in step 6.1 to ensure that the observation period is synchronized with the behavior adjustment.

[0060] Step 6.3: Calculate the difference in trend slope before and after the intervention and quantify the effect: To objectively measure the intervention effect, the system calculates the change in the trend slope before and after the intervention. First, for... Perform a quadratic fitting to obtain the post-intervention acceleration. ; and before the warning Comparison, defining the effect index ; in, The coefficient of the quadratic term of the early warning trend. This represents the coefficient of the quadratic term of the trend after intervention. If... This indicates that the deteriorating trend has been contained or reversed; if This indicates that the situation has further deteriorated; if This indicates no significant change. The principle behind this formula is to quantify the net effect of intervention through the difference in acceleration, avoiding the randomness that comes from relying solely on a single point value. For example, even if blood glucose levels do not completely return to the normal range, as long as the rate of increase slows down ( This can still be considered partially valid. The calculation directly references the result extracted in step 6.2. With step three .

[0061] Step 6.4: Update attribution confidence and revise future risk assessment logic: Based on the effectiveness index The system updates the attribution confidence level for this type of risk event. If... (Empirical threshold), and if the user confirms the intervention, then the initial attribution is deemed correct, and the priority for identifying similar patterns is increased; if If a user marks a signal as a "false alarm," its weight is reduced to avoid repeated misjudgments. For example, if intervention is ineffective after multiple blood glucose warnings, the system is more likely to attribute similar trends to "unknown causes" rather than "dietary factors" in the future, guiding earlier medical advice. This mechanism achieves a logical closed loop of reasoning from the outcome to the premise, allowing the assessment path to be continuously optimized with experience. This process directly relies on step 6.3. Values ​​should be updated based on objective criteria.

[0062] Step 7: Construct a long-term health evolution map and support cross-cycle comparisons After completing the closed-loop processing of a single risk event, the system integrates all interim results into a long-term health evolution map. This map not only records the occurrence, development, and outcome of each risk event, but also supports trend comparisons across time periods, helping to identify early cumulative changes in chronic diseases and compensating for the limitations of short-term analysis.

[0063] Step 7.1: Archive the data from the completed risk assessment cycle: After each risk event is resolved, the system archives all key data from that period in a structured manner. The archived content includes: the initial baseline. Exception event set Trend analysis results Attribution conclusions, warning levels, user responses, and post-intervention trends and effectiveness index These data are stored chronologically in an encrypted database, forming the individual's "health event profile." Each record is accompanied by a unique event ID, enabling rapid retrieval. This process references the feedback verification results completed in step six to ensure the integrity of the archived information and the certainty of the conclusions.

[0064] Step 7.2: Extract periodic risk events and analyze recurrence patterns: After accumulating at least three similar risk events, the system initiates periodic analysis. For example, if a user experiences four episodes of elevated blood glucose levels within the past six months, the system extracts the trigger time, duration, attribution category, and intervention effect for each event to construct a recurrence time series. By calculating the time interval between adjacent events. Analyze its changing trend. If A decreasing trend (e.g., from 90 days → 60 days → 45 days) suggests an increased risk of recurrence and may indicate the onset of a pre-disease state. This analysis relies on the event timestamps archived in step 7.1 to ensure sequence accuracy.

[0065] Step 7.3: Calculate the long-term health degradation index and visualize the evolution path: To quantify long-term changes in an individual's overall health status, a health degradation index is systematically defined. ; in, This represents the total number of risk events handled in the past year. For the first Trend risk score for this event Its theoretical maximum value (e.g., 1.5). This is the effect index of the event. To prevent division by zero for the smoothing constant. The principle behind this formula lies in considering both the intensity of risk and the difficulty of intervention: high... This indicates a serious problem, low This indicates that the problem is difficult to correct, and the combination of these two factors reflects a decline in system stability. A higher value indicates a more significant decline in health. This index is calculated based on archived data from steps 7.1 and 6.3 to ensure long-term sustainability.

[0066] Step 7.4: Generate a cross-cycle health report and provide preventative recommendations: Based on health degradation index Based on relapse pattern analysis, the system generates an annual health evolution report. The report displays trends in key indicators, distribution of risk events, and changes in intervention effectiveness in chart format. If the value rises for two consecutive years, or the recurrence interval of a certain type of event shortens by more than 50%, then it is recommended that the user undergo specific screening (such as a glucose tolerance test). This report not only reviews the past but also provides preventative recommendations based on evolutionary patterns, such as "It is recommended to monitor blood glucose trends quarterly to avoid entering peak seasons." This output directly references step 7.3. The value and the recurrence pattern in step 7.2 enable the transformation from data to insights.

[0067] This embodiment achieves a proactive and personalized assessment of health risks through continuous data collection, multi-dimensional correlation analysis, dynamic weight adjustment, and closed-loop feedback verification. Compared to static models, its core advantage lies in its ability to capture gradual changes and sudden fluctuations in an individual's condition, promptly identify potential crises, and use intervention results to feed back into the assessment logic, forming a sustainable and evolving health protection mechanism. The entire process is data-driven, time-based, and individual-centered, constructing a complete intelligent assessment path from perception to response, and from early warning to verification.

[0068] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent risk assessment in disease detection, characterized in that, Includes the following steps: Collect an individual's date of birth, gender, family history of genetic diseases, and records of past major illnesses to generate a unique identifier; Physiological measurements taken during the past 6 months while the individual was in a healthy state were retrieved, and the mean and standard deviation of each indicator were calculated. The normal fluctuation range for an individual was defined as the mean ± 1.5 times the standard deviation. By integrating behavioral data from 30 consecutive days, we analyzed the temporal distribution of average daily steps, sleep duration, and dietary patterns to extract characteristics of daily activities. Based on the information corresponding to the identity identifier, the normal fluctuation range and the characteristics of daily activities, the initial risk weights for different disease categories are allocated according to a preset ratio; By connecting to real-time data streams from wearable devices and environmental sensors, current physiological indicators are compared with the individual's normal fluctuation range, deviations are marked, and unexplained abnormal signals are filtered in conjunction with environmental parameters.

2. The intelligent risk assessment method for disease detection according to claim 1, characterized in that, The extraction of features from daily activities includes: Calculate the mean and standard deviation for each behavioral indicator; The ratio of standard deviation to mean is used to measure the stability of behavioral rhythms; Identify behavioral patterns that deviate from the mean by more than 20% over seven consecutive days; Behaviors with low stability and persistent deviations are labeled as rhythmic perturbations.

3. The intelligent risk assessment method for disease detection according to claim 1, characterized in that, The allocation of initial risk weights for different disease categories according to a preset ratio includes: Set the disease predisposition corresponding to a family history of genetic diseases as the first vector; The relative distance between each physiological indicator and the midpoint of its normal fluctuation range is set as the second vector; Transform the stability of behavioral rhythms into a third vector; The initial risk weight distribution is obtained by weighted summation of the three vectors.

4. The intelligent risk assessment method for disease detection according to claim 1, characterized in that, The marker deviations, combined with environmental parameters, are used to filter unexplained anomalous signals, including: The physiological indicators are compared with the individual's normal fluctuation range every 5 minutes; When the current value exceeds the individual's normal fluctuation range, it is marked as an initial deviation; Simultaneously acquire ambient temperature, humidity, and air quality data. If the environmental parameters are within a known interference range, reduce the risk level of the deviation.

5. The intelligent risk assessment method for disease detection according to claim 4, characterized in that, The method of marking deviation terms and filtering unexplained anomalous signals in conjunction with environmental parameters also includes: For the unexplained anomalous signals that were screened out, time-series segments were formed by tracing back the measurement records of the past 7 days. Apply a 24-hour sliding window to the time series segment, calculate the difference between the final value and the initial value within the window, divide by the time span, and obtain the local rate of change. The trend curve is obtained by least-squares fitting of the time series segments using a quadratic polynomial; The predicted value for the next 12 hours is extrapolated from the trend curve and compared with the upper limit of the individual's normal fluctuation range to determine whether there is a trend risk.

6. The intelligent risk assessment method for disease detection according to claim 5, characterized in that, The process of using a quadratic polynomial to perform least-squares fitting on time segments to obtain trend curves includes: The quadratic polynomial is set to be a linear combination of the time square term, the time linear term, and the constant term. The multipliers of the three terms are adjusted to minimize the sum of squared errors between the fitted result and the actual measured value. The term with the largest multiplier is retained as the dominant trend feature, and the dominant trend feature is used to determine whether the change in the indicator is an acceleration or a slowdown.

7. The intelligent risk assessment method for disease detection according to claim 5, characterized in that, The method of extrapolating the predicted value for the next 12 hours based on the trend curve and comparing it with the upper limit of the individual's normal fluctuation range to determine whether there is a trend risk includes: Extract dietary records, exercise duration, sleep quality, and indoor / outdoor temperature data from the same period. Calculate the ratio of the covariance to the standard deviation of each data point and abnormal physiological indicator over 7 days; Data with ratios greater than 0.5 are marked as significantly relevant variables; The presence or absence of significantly relevant variables determines whether the risk originates from lifestyle or underlying pathological processes.

8. The intelligent risk assessment method for disease detection according to claim 7, characterized in that, The method of determining whether a risk originates from lifestyle or an underlying pathological process based on the presence or absence of significantly relevant variables includes: If the attribution result is lifestyle influence, then the current disease category weight remains unchanged; If the attribution result is an underlying pathological process, the weight of the corresponding disease category will be increased by 10% to 15%. Other category weights are compressed proportionally to maintain a total of 100%; Three-level early warning thresholds are set based on the updated weights and trend risk scores.

9. The intelligent risk assessment method for disease detection according to claim 8, characterized in that, The three-level early warning threshold is set based on the updated weights and trend risk scores, including: Health advice is sent when the trend risk score is greater than 0.5; When the trend risk score is greater than 0.8, a medical examination is recommended. When the trend risk score is greater than 1.2, notify the emergency contact person. Record the warning content, trigger time, and supporting data, and enter the pending feedback state.

10. A disease detection intelligent risk assessment system for implementing the method as described in any one of claims 1-9, characterized in that, include: The data acquisition unit is used to obtain an individual's date of birth, gender, family history of genetic diseases, records of past major illnesses, and physiological measurements of their health status over the past 6 months. The interval generation unit is used to calculate the mean and standard deviation based on physiological measurements, and to define the normal fluctuation range of an individual with the mean ± 1.5 times the standard deviation. The feature extraction unit is used to integrate behavioral data from 30 consecutive days, analyze the temporal distribution of average daily steps, sleep duration and dietary regularity, and extract features of daily activities. The weighting allocation unit is used to combine the information corresponding to the identity identifier, the individual's normal fluctuation range and daily activity characteristics, and allocate the initial risk weights of different disease categories according to a preset ratio. The anomaly screening unit is used to access the real-time data streams from wearable devices and environmental sensors, compare the current physiological indicators with the individual's normal fluctuation range, mark deviations, and filter out unexplained abnormal signals in conjunction with environmental parameters.

Images