Depression assessment titration optimization method and system based on professional labels

Abstract

The invention provides a depression assessment titration optimization method and system based on a professional label, and the method comprises the steps: collecting the heart rate variability, skin conductance, ambient light intensity, wrist body movement and voice data of a user according to a fixed period through a wearable device, inputting the data of each period into a unified assessment model trained by a professional medical label, the time sequence fusion network processes and outputs a depression state evaluation result containing the total score and the contribution value of each dimension; generating a standardized evaluation conclusion based on the multi-cycle prediction result, and accurately associating the original data with the standardized conclusion; and continuously optimizing model parameters according to an association result by utilizing an incremental learning technology, gradually generating a personalized evaluation model adaptive to individual characteristics of the user, and realizing progressive improvement of evaluation precision. According to the invention, the accuracy and individuation degree of dynamic monitoring of the depression state are improved.

Description

Technical Field

[0001] The present application relates to the technical field of professional labels, and in particular to a depression assessment titration optimization method and system based on professional labels. Background Art

[0002] Current mental health assessments of depression require a balance of objectivity, real-time performance, and personalization. Traditional scale-based assessments suffer from subjectivity and long evaluation cycles. Objective assessment methods based on physiological data require the integration and analysis of heterogeneous data from multiple sources, while also requiring clinically reliable and individually adaptable results. There is an urgent need for intelligent analysis technologies that can continuously track changes in a user's physiological state and dynamically optimize assessment models.

[0003] Existing technologies use static assessment models based on machine learning algorithms. This model is trained by collecting physiological data such as electrocardiogram (ECG) and electrical skin conductance during a single period of time, combined with scores on psychological scales. Once deployed, the model maintains fixed parameters, periodically receives new user data, and outputs assessment results. Depression risk levels are determined by setting fixed thresholds.

[0004] Static models struggle to adapt to the long-term evolution of individual physiological characteristics. As a user's physiological state changes, assessment accuracy gradually declines. Model optimization requires re-collecting large amounts of annotated data and conducting comprehensive training, causing the assessment to lag behind actual changes in the user's state. Furthermore, single-point data analysis ignores the dynamic correlations between physiological indicators, compromising the comprehensiveness of the assessment dimensions. Summary of the Invention

[0005] The present application provides a depression assessment titration optimization method and system based on professional labels to solve the problems of low accuracy and low personalization in dynamic monitoring of depression status in the prior art.

[0006] In a first aspect, the present application provides a depression assessment titration optimization method based on professional labels, comprising: Collecting multimodal data of the user according to a preset period, wherein the multimodal data includes: heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data, and voice data; Inputting the multimodal data within the same period into a unified depression assessment model trained with professional labels, processing the multimodal data through a temporal fusion network in the depression assessment model, and outputting a predicted depression state assessment result, the depression state assessment result including a total depression state score and a contribution value of each physiological characteristic dimension; Generating a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and associating the multimodal data within a preset period with the standardized depression assessment result; Based on the association results, the parameters of the unified depression assessment model are optimized through incremental learning to generate a personalized depression assessment model for the user, thereby achieving progressive titration optimization of depression assessment.

[0007] Optionally, the step of optimizing the parameters of the unified depression assessment model by incremental learning based on the association result to generate a personalized depression assessment model for the user includes: Compare the standardized depression assessment results in the correlation results with the predicted depression state assessment results, and calculate the assessment deviation value of each physiological characteristic dimension based on the comparison results; Determining a parameter adjustment direction instruction for a corresponding feature extraction path in the time series fusion network according to the evaluation deviation value; Through incremental learning, while retaining the original parameter structure of the unified depression assessment model, the neuron weights involved in the parameter adjustment direction instructions are incrementally updated to generate a personalized depression assessment model that adapts to the user's individual physiological response pattern.

[0008] Optionally, determining the parameter adjustment direction instruction of the corresponding feature extraction path in the time series fusion network according to the evaluation deviation value includes: Analyzing the distribution of the evaluation deviation values ​​in each physiological characteristic dimension; Determining, based on the distribution, a specific dimension whose deviation contribution value exceeds a preset deviation contribution threshold; According to the correspondence between each physiological feature dimension and the feature extraction path in the time series fusion network, determining the neural network processing path associated with the specific dimension as the target feature extraction path; Based on the size of the evaluation deviation value of the specific dimension, a parameter adjustment direction instruction for the target feature extraction path is generated.

[0009] Optionally, the generating of a parameter adjustment direction instruction for the target feature extraction path based on the evaluation deviation value of the specific dimension includes: Matching the assessment deviation value of the specific dimension with a preset clinical grading threshold set; If the matching result indicates that the assessment deviation value is greater than the clinical moderate depression threshold in the clinical grading threshold set, generating a positive instruction for enhancing the emotion-related feature capture capability in the target feature extraction path; Alternatively, if the matching result indicates that the assessment deviation value is less than the clinical mild depression threshold value in the clinical grading threshold set, generating a negative instruction for suppressing noise sensitivity in the target feature extraction path; According to the deviation amplitude ratio between the evaluation deviation value and the preset clinical grading threshold, the intensity level of the positive instruction or the negative instruction is dynamically set to form a parameter adjustment direction instruction.

[0010] Optionally, processing the multimodal data through a temporal fusion network in the depression assessment model to output a predicted depression state assessment result includes: The time series fusion network performs parallel time series scanning on the heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data, and voice data within the same period, and extracts dynamic change features related to emotional fluctuations from each modal data; establishing a dynamic response relationship between different modal data based on the time synchronization of the dynamic change characteristics; Based on the dynamic response relationship and the preset professional label mapping rules, calculating the independent contribution value of each physiological characteristic dimension to the depressive state; Aggregate the contribution values ​​of each physiological characteristic dimension to generate a total score of depression state; The contribution value of each physiological characteristic dimension and the total score of the depression state are combined to form a depression state assessment result.

[0011] Optionally, generating a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and associating the multimodal data within a preset period with the standardized depression assessment result, includes: Arrange the predicted depression status assessment results corresponding to all periods in chronological order to form a user depression status evolution sequence; Generate a standardized depression assessment result based on the fluctuation trend of the user's depression state evolution sequence and in combination with preset clinical assessment criteria; The multimodal data within a complete cycle corresponding to the standardized depression assessment result is bound one-to-one with the standardized depression assessment result according to the data collection timestamp to form an associated data set, which is the associated result.

[0012] Optionally, the incremental updating of the neuron weights involved in the parameter adjustment direction instruction to generate a personalized depression assessment model adapted to the user's individual physiological response pattern includes: Integrate the associated dataset of the preset period with the associated dataset accumulated during the historical optimization period, and calculate the model parameter update amount based on the incremental learning method; Iteratively optimizing the unified depression assessment model according to the model parameter update amount to generate a personalized depression assessment model, and applying the personalized depression assessment model to the multimodal data analysis of the next cycle to obtain a depression assessment result; The above steps are repeated to gradually adapt the personalized depression assessment model to the user's unique physiological behavior, thereby achieving a gradual improvement in assessment accuracy.

[0013] In a second aspect, the present application provides a depression assessment titration optimization system based on professional labels, comprising: An acquisition module is used to collect multimodal data of the user according to a preset period, wherein the multimodal data includes: heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data, and voice data; An input module is configured to input the multimodal data within the same period into a unified depression assessment model trained with professional labels, process the multimodal data through a temporal fusion network in the depression assessment model, and output a predicted depression state assessment result, wherein the depression state assessment result includes a total depression state score and a contribution value of each physiological characteristic dimension; A generation module, configured to generate a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and to associate the multimodal data within a preset period with the standardized depression assessment result; The optimization module is used to optimize the parameters of the unified depression assessment model through incremental learning based on the association results to generate a personalized depression assessment model for the user and realize progressive titration optimization of depression assessment.

[0014] In a third aspect, the present application provides a computing device comprising a processor and a memory, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute a depression assessment titration optimization method based on professional labels as described in any one of the first aspects.

[0015] In a fourth aspect, the present application provides a computer storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement a depression assessment titration optimization method based on professional labels as described in any one of the first aspects.

[0016] In the present application, a depression assessment titration optimization method based on professional labels is provided, which includes: collecting multimodal data of a user according to a preset period, the multimodal data including: heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data and voice data; inputting the multimodal data within the same period into a unified depression assessment model trained with professional labels, processing the multimodal data through a time series fusion network in the depression assessment model, and outputting a predicted depression state assessment result, the depression state assessment result including a total depression state score and a contribution value of each physiological characteristic dimension; generating a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and associating the multimodal data within the preset period with the standardized depression assessment result; based on the association result, optimizing the parameters of the unified depression assessment model through incremental learning to generate a personalized depression assessment model for the user, thereby achieving progressive titration optimization of depression assessment.

[0017] The technical solution provided by this application has the following beneficial effects: This application realizes the continuous acquisition of depression status assessment data, establishes a stable and reliable data source for dynamic monitoring, and ensures the continuity and comprehensiveness of the assessment. Through the time series fusion network, it comprehensively analyzes various types of physiological and behavioral data, captures the dynamic correlation between characteristics of different dimensions, and improves the scientificity and accuracy of the assessment results. It provides an intuitive and quantitative depression status score, and at the same time shows the specific impact of each physiological indicator, providing a basis for precise intervention. The model prediction results are converted into assessment conclusions that meet clinical standards, and a corresponding relationship between the original data and the medical diagnosis is established to ensure the medical credibility of the assessment. The model continuously adapts to changes in the user's physiological characteristics, gradually improves the assessment accuracy, and realizes personalized depression status tracking that varies from person to person.

[0018] Furthermore, this application also locates the assessment differences of each physiological dimension by comparing the deviation values ​​of the standardized assessment results and the predicted results, and specifically determines the adjustment direction of the neural network feature extraction path. It uses incremental learning to update the parameters of only the deviation-related paths while maintaining the stability of the overall structure of the model, and finally generates a depression assessment model that meets the individual characteristics of the user.

[0019] In addition, this method realizes the dynamic optimization of the depression assessment model, enabling the model to autonomously adjust as the user's physiological state changes. It not only retains existing knowledge but also adapts to the evolution of individual characteristics, improves the personalization and accuracy of the assessment results, and ensures the efficiency and stability of model optimization.

[0020] These and other aspects of the present application will become more readily apparent from the description of the following embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0022] Figure 1 A flowchart of a depression assessment titration optimization method based on professional labels provided in an embodiment of the present application; Figure 2 A schematic diagram of the structure of a depression assessment titration optimization system based on professional tags provided in an embodiment of the present application; Figure 3 A schematic diagram of the structure of a computing device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0024] In some of the processes described in the specification and claims of this application and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this document or may be executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish between different operations, and the serial numbers themselves do not represent any order of execution. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to being different types.

[0025] Current depression assessment technologies based on static models have significant shortcomings. These methods employ fixed-parameter assessment models that cannot be adjusted once deployed, making them difficult to adapt to physiological differences between different users or within the same user over time. As a user's depressive symptoms change, the deviation between the assessment results output by the static model and the actual state gradually increases, and this deviation cannot be corrected through simple parameter adjustments. Furthermore, these models typically only analyze data from a single point in time, ignoring the dynamic correlations between physiological indicators, resulting in a lack of comprehensiveness and individual adaptability in the assessment results.

[0026] In response to the above problems, this application proposes a depression assessment titration optimization method based on professional labels. The core of this method is to dynamically optimize the model parameters by periodically collecting the user's multimodal physiological data and combining it with the standardized evaluation results of professional medical institutions. Specifically, after each round of evaluation cycle, the system compares the predicted results with the professional labels, calculates the deviation values ​​of each physiological dimension, and only makes incremental adjustments to the relevant neural network paths. This method breaks through the limitations of static models and enables the evaluation model to be continuously optimized as the user's status changes, which not only ensures the accuracy of medical evaluation, but also realizes personalized analysis based on individual differences. Through this progressive adjustment strategy, the system can accurately capture the evolution trend of the user's physiological characteristics, fundamentally solving the problems of evaluation lag and insufficient individual adaptation in the existing technology, and improving the reliability and practicality of depression status monitoring.

[0027] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of this application.

[0028] Figure 1 A flowchart of a depression assessment titration optimization method based on professional labels provided in an embodiment of the present application is shown in FIG. Figure 1 As shown, the method includes: Step 101: Collect multimodal data of the user according to a preset period, wherein the multimodal data includes: heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data, and voice data.

[0029] In step 101, the preset period refers to a fixed time interval set by the system for regularly collecting user data to ensure continuous evaluation. Heart rate variability data reflects changes in heartbeat intervals, reflecting autonomic nervous system regulation. Skin conductance data displays sweat gland activity, indicating the level of emotional arousal. Ambient light intensity data records changes in light intensity, indicating circadian rhythm status. Wrist motion data captures the amplitude and frequency of wrist movements, reflecting behavioral activity levels. Voice data analyzes acoustic characteristics such as intonation and speech rate through timed recording.

[0030] In an embodiment of the present application, the system automatically collects data on a weekly basis through the smart device worn by the user. The optical sensor continuously monitors changes in ambient light intensity, the accelerometer records wrist movement, the electrocardiogram module detects fluctuations in heartbeat intervals, the galvanic skin sensor measures changes in hand conductivity, and the microphone periodically records voice clips. After denoising, the data from various sensors is uniformly stored in a local database by timestamp, forming a periodic data set containing complete physiological and behavioral characteristics. When the preset 7-day period node is reached, the system automatically packages all valid data within that period and prepares it for transmission to the analysis module.

[0031] For example, user A completes their first week of monitoring wearing a smart bracelet. The device collects 672 sets of heart rate variability data (once per minute), 2016 sets of skin conductance data (once every five minutes), 10,080 sets of ambient light intensity data (once per minute), 30,240 sets of wrist motion data (once every 20 seconds), and 84 segments of voice data (once every two hours). All data is initially filtered by the device's built-in processor to remove outliers caused by improper wear (e.g., five consecutive heart rate intervals exceeding 200ms). The remaining valid data, accounting for 98.5%, is sorted by collection time to form the first week's dataset.

[0032] Step 102: The multimodal data within the same period is input into a unified depression assessment model trained with professional labels, the multimodal data is processed by the time series fusion network in the depression assessment model, and a predicted depression state assessment result is output, wherein the depression state assessment result includes a total depression state score and a contribution value of each physiological characteristic dimension.

[0033] In step 102, professional labels refer to professional medical and psychological assessment labels. These are authoritative diagnostic results generated by medical institutions after professionally assessing a patient's depression using standardized psychological assessment tools (such as the PHQ-9 scale and the HAMD scale). These labels contain depression severity grading and symptom descriptions that conform to clinical diagnostic criteria. They serve as the gold standard for model training and optimization, ensuring the medical credibility of the assessment results. The time series fusion network is a specially designed deep learning architecture. The predicted depression assessment results are quantitative conclusions derived by the model through analysis of the user's multimodal physiological data. They include an overall score reflecting the overall level of depression (the total depression score) and multiple detailed scores (the contribution values ​​of each physiological characteristic dimension) that indicate the impact of each physiological indicator. The overall score is used to determine the severity of depression, while the contribution values ​​of each dimension specifically indicate the impact of different physiological characteristics such as heart rate and voice on the current depression state, providing a precise reference for subsequent personalized intervention.

[0034] In this embodiment of the present application, after the system inputs the periodic dataset into the model, each physiological data item enters the corresponding feature extraction branch. The heart rate data is processed through the fluctuation analysis module to extract emotion-related feature segments, and the voice data is processed through the voiceprint recognition module to extract intonation variation features. All features are aligned along the time axis and sent to the association analysis layer. This layer calculates the co-occurrence probability of different features in the same time period to establish dynamic association rules, such as heart rate acceleration accompanied by voice tremor. The final contribution value calculation unit converts the feature association strength into contribution scores for each dimension based on the clinical standards defined by the professional label, and outputs the total depression state score after aggregation.

[0035] For example, after inputting user A's first week of data into the model, the system discovered a correlation between heart rate fluctuations (interval variations of up to 25ms) between 3:00 AM and 5:00 AM and the flatness of voice (a 40% decrease in intonation) in a speech clip from the same time period. Based on pre-set rules, this combined pattern corresponds to a depression contribution score of 3. Combined with analysis of data from other time periods, the final first-week assessment results were output: a total score of 14, including 3.5 for heart rate, 2.8 for speech, 3.2 for body movement, 2.5 for light intensity, and 2 for skin conductance.

[0036] Step 103: Based on the predicted depression state assessment results corresponding to all periods, a standardized depression assessment result of the user is generated, and the multimodal data in the preset period is associated with the standardized depression assessment result.

[0037] In step 103, the standardized depression assessment result refers to a graded conclusion that meets the hospital's diagnostic criteria, such as "mild depression" or "moderate depression."

[0038] In this embodiment, the system aggregates a user's prediction results over multiple consecutive weeks, analyzing trends in the overall score and fluctuations in the contributions of each dimension. When the user visits the hospital for a scheduled follow-up appointment, they enter their professional scale score into the system, and the algorithm automatically matches the clinical diagnosis with recent physiological data. During this matching process, the system establishes rules, such as "a light intensity contribution value consistently above 2.5 points corresponds to a doctor-diagnosed sleep disorder," ultimately generating a time-stamped, linked dataset.

[0039] For example, after User A completed four weeks of monitoring, the system recorded a gradual decrease in their total score from 14 to 9, with the light intensity dimension contributing consistently above 2.5 points. A follow-up visit to the hospital in week 4 revealed a PHQ-9 score of 10 (mild depression), with the doctor specifically noting sleep problems. Based on this, the system established an association rule linking "abnormal light intensity data to sleep disturbances to depression scores" and stored all raw data from week 4, binding it to this clinical conclusion.

[0040] Step 104: Based on the association results, the parameters of the unified depression assessment model are optimized through incremental learning to generate a personalized depression assessment model for the user, thereby achieving progressive titration optimization of depression assessment.

[0041] In step 104, incremental learning refers to a technique that updates the model based on newly discovered feature relationships while retaining existing knowledge. The personalized depression assessment model is a proprietary version optimized for user-specific data.

[0042] In this embodiment, the system compares the latest associated dataset with historical data to identify individual patterns, such as User A's light intensity sensitivity. During model optimization, most network parameters are first frozen, leaving only the weight adjustment permission for the light intensity processing pathway. Through small-scale training, this pathway is made more sensitive to light characteristics while ensuring that the assessment capabilities of other dimensions remain unaffected. The updated model will more accurately capture User A's light-related depressive symptoms in the next cycle.

[0043] For example, the system optimized the model's light analysis module specifically to target user A's light sensitivity. After three incremental updates, the new model's assessment of user A narrowed the gap between the hospital's diagnosis and the initial 2 points to 0.5 points. The accuracy of predicting depression was particularly improved during periods of sudden changes in light intensity.

[0044] This method periodically collects multidimensional physiological data and integrates it with professional medical assessments to enable dynamic monitoring and precise analysis of depressive states. The model automatically discovers individual patterns and continuously optimizes itself, ensuring both medical accuracy and adaptability to the unique characteristics of each user. The entire system operates without human intervention, forming a complete closed loop from data collection to model updates, providing reliable technical support for the long-term management of depressive symptoms.

[0045] To address the problem of insufficient personalized adaptation of depression assessment models, in some embodiments, step 104: optimizing the parameters of the unified depression assessment model through incremental learning based on the association results to generate a personalized depression assessment model for the user, includes: Step 201: Compare the standardized depression assessment result in the correlation result with the predicted depression state assessment result, and calculate the assessment deviation value of each physiological characteristic dimension based on the comparison result.

[0046] In step 201, the evaluation deviation value refers to the difference between the model prediction result and the professional medical evaluation result in each physiological dimension, reflecting the degree of deviation in the model's understanding of specific physiological characteristics.

[0047] In this example, the system first obtains the hospital's standard depression score and the model's prediction results, then compares the differences between the two scores in dimensions such as heart rate and speech. For the heart rate dimension, the difference between the predicted contribution value and the actual medical assessment is calculated. For the speech dimension, the degree of deviation between the intonation feature score and the clinical conclusion is analyzed. Finally, an evaluation report containing the specific deviation values ​​for each dimension is output.

[0048] Step 202: Determine a parameter adjustment direction instruction for a corresponding feature extraction path in the temporal fusion network according to the evaluation deviation value.

[0049] In step 202, the feature extraction pathway refers to the neural network processing channel within the depression assessment model specifically responsible for identifying emotion-related features from a certain type of physiological data. These pathways are automatically formed during the initial training of the model based on the correspondence between professional medical labels and multimodal data. For example, the heart rate data processing pathway specifically analyzes heartbeat interval fluctuation patterns, while the speech processing pathway focuses on identifying intonation changes. Each pathway optimizes feature extraction methods for a specific data type and can be individually adjusted during model optimization without affecting other data processing functions. Parameter adjustment direction instructions include two types: enhancement and suppression. The adjustment intensity is determined by the degree to which the deviation value exceeds the clinical threshold. Each instruction corresponds to a specific neural network pathway, such as the heart rate processing pathway or the speech analysis pathway.

[0050] In this embodiment, after reading the deviation report, the system compares the heart rate deviation value with a preset threshold. If it exceeds the moderate depression threshold, it generates an instruction to enhance the heart rate feature extraction capability. It also detects the voice deviation value. If it falls below the mild threshold, it generates an instruction to suppress the voice path sensitivity. All instructions are marked with the specific intensity level of action.

[0051] Step 203: By incremental learning, while retaining the original parameter structure of the unified depression assessment model, the neuron weights involved in the parameter adjustment direction instruction are incrementally updated to generate a personalized depression assessment model adapted to the user's individual physiological response pattern.

[0052] In step 203, incremental updates involve making small parameter adjustments to specific neural network nodes while maintaining the model's basic functionality. The update process retains proven feature recognition patterns and only corrects any processing logic that deviates.

[0053] In this embodiment of the present application, upon receiving an adjustment command, the model first targets network components unrelated to heart rate analysis. It then locates the specific neuron group responsible for extracting heart rate fluctuation features and increases its connection weights proportionally to the command strength. The speech processing component selectively lowers certain filtering thresholds. The updated model retains the original framework but optimizes individual functional modules.

[0054] Here's a specific example: After user A completed the fourth week of monitoring, the system compared the model-predicted depression assessment results (total score of 9 points, including 2.8 points for the light intensity dimension) with the standardized depression assessment results issued by the hospital (PHQ-9 score of 10 points, with sleep disorders marked). First, the assessment deviation values ​​of each dimension were calculated, among which the light intensity dimension deviation value was +0.8 points (2.8 points model prediction value minus 2 points standard value). This value was calculated using the formula ΔS=S_model-S_clinic, where ΔS represents the deviation value, S_model represents the model prediction contribution value, and S_clinic represents the clinical standard value. According to the preset rules, when the light intensity dimension deviation value exceeds the clinical threshold of 0.5 points, the system determines that the sensitivity of the illumination feature extraction path needs to be enhanced and generates an adjustment instruction with an intensity level of 3 ( Intensity level = deviation value 0.8 points / adjustment unit 0.25 points, rounded up); during the incremental learning process, the model keeps other processing paths such as heart rate and voice unchanged, and only enhances and adjusts the weights of the three groups of neurons in the light analysis module responsible for identifying abnormal light patterns, updating the original weight value Wi to Wi'=Wi+α·L, where α=0.1 is the learning rate and L=3 is the intensity level. After this optimization, the accuracy of identifying the depressive state of user A during the abnormal light intensity data period in the fifth week (light intensity below 50 lux between 1 and 4 a.m.) was significantly improved. The contribution value of the light intensity dimension output by the model was adjusted to 2.1 points, and the deviation from the hospital follow-up PHQ-9 score of 9 points was reduced to 0.3 points. The evaluation results of other dimensions remained stable, verifying the accuracy and safety of incremental optimization.

[0055] In the embodiments of the present application, the method achieves precise and progressive optimization of the depression assessment model, enabling the model to continuously track the user's physiological changes and improve itself, ensuring that the assessment results meet medical standards and meet personalized needs, greatly improving the reliability and practicality of long-term monitoring.

[0056] To further improve the accuracy of depression assessment model optimization, in some embodiments, step 202: determining the parameter adjustment direction instruction of the corresponding feature extraction path in the time series fusion network based on the assessment deviation value includes: Step 301: Analyze the distribution of the evaluation deviation value in each physiological characteristic dimension.

[0057] In step 301 , the distribution refers to the proportion and distribution characteristics of the prediction deviations of different physiological dimensions (such as heart rate, voice, etc.) in the overall deviation, reflecting the degree and pattern of influence of each dimension on the total deviation.

[0058] In this embodiment, the system first normalizes the calculated deviation values ​​for each dimension, analyzes which dimensions have consistently high deviations and which dimensions have large fluctuations, and forms a distribution map containing the deviation weights and change trends for each dimension. This map shows that the deviation in the heart rate dimension accounts for 35%, the speech dimension accounts for 25%, the light intensity dimension accounts for 20%, the body movement dimension accounts for 15%, and the skin conductance dimension accounts for 5%.

[0059] Step 302: According to the distribution, determine a specific dimension whose deviation contribution value exceeds a preset deviation contribution threshold.

[0060] In step 302, the deviation contribution is a quantitative indicator derived from the decomposition of the assessment deviation values ​​across various physiological characteristic dimensions. It represents the degree of influence of each dimension on the overall assessment deviation. Its value is directly derived from the normalization of the assessment deviation values ​​corresponding to each dimension. The preset deviation contribution threshold is a minimum impact value set based on clinical experience. Specific dimensions include heart rate variability, skin conductance, ambient light intensity, wrist motion, or speech.

[0061] In an embodiment of the present application, the system calculates the contribution of each dimension based on the distribution map, among which the contribution of the heart rate dimension reaches 0.35, exceeding the preset threshold of 0.3, and the voice dimension of 0.25 is close to the threshold. Therefore, the heart rate dimension is determined as the primary adjustment target and the voice dimension is determined as the secondary adjustment target.

[0062] Step 303: According to the correspondence between each physiological feature dimension and the feature extraction path in the temporal fusion network, the neural network processing path associated with the specific dimension is determined as the target feature extraction path.

[0063] In step 303, the corresponding relationship refers to the mapping rules between the pre-established physiological dimensions and the neural network processing channels, ensuring that each physiological data type has a dedicated optimization channel for processing. The neural network processing path specifically refers to the substructure branch in the time series fusion network that specifically processes a specific physiological feature dimension. These paths together constitute the overall architecture of the time series fusion network, and each path specifically processes the feature extraction and conversion tasks of the corresponding dimension. The target feature extraction path refers to the subnetwork channel in the time series fusion neural network that is specifically responsible for processing specific physiological dimension data. These paths establish a fixed corresponding relationship with the specific physiological data type during the initial design of the model. For example, heart rate variability data corresponds to the autonomic nerve feature extraction path, and speech data corresponds to the emotional speech analysis path. When it is determined that the evaluation effect of a certain physiological dimension needs to be optimized, the system will lock the dedicated processing channel corresponding to the dimension as the adjustment target to ensure that the model optimization process is both accurate and does not interfere with the normal operation of other unrelated functions.

[0064] In an embodiment of the present application, the system queries a preset mapping table to determine that the heart rate dimension corresponds to the autonomic nervous feature extraction path, and the speech dimension corresponds to the emotional speech analysis path, and marks these two paths as the target paths for this optimization.

[0065] Step 304: Based on the evaluation deviation value of the specific dimension, generate a parameter adjustment direction instruction for the target feature extraction path.

[0066] In this embodiment of the present application, for a positive deviation of 0.35 in the heart rate dimension, the system generates an instruction to enhance the sensitivity of the autonomic nervous system pathway, with an intensity level of 3; for a negative deviation of 0.25 in the speech dimension, an instruction to suppress overfitting of the speech pathway is generated, with an intensity level of 2. All instructions are associated with specific target pathways and adjustment amplitudes.

[0067] Here's a specific example: After user A's fifth week of monitoring, the system first analyzed the distribution of deviations in each dimension assessment. It found that the deviation in the light intensity dimension remained at +0.7 points (the difference between the model's predicted value of 2.1 and the baseline value of 1.4 points), accounting for 46% of the total deviation and far exceeding the 30% contribution threshold. Meanwhile, the deviation in the heart rate dimension had dropped to +0.2 points, and the deviation in the speech dimension had dropped to -0.3 points. Based on the pre-set dimension-path mapping table, the light intensity dimension was associated with the illumination feature extraction path, and the speech dimension with the affective speech analysis path. For the persistently high +0.7-point deviation in the light intensity dimension (exceeding the moderate depression threshold of 0.5 points but below the severe depression threshold of 1.0 points), the system generated an enhancement command of intensity level 2 (intensity level = deviation value 0.7 points / adjustment unit 0.35 points, rounded to the nearest whole number). Simultaneously, for the -0.3-point deviation in the speech dimension (below the mild depression threshold of -0.2 points), a suppression command of intensity level 1 was generated.

[0068] In the embodiment of the present application, this method realizes the precise and targeted optimization of the depression assessment model parameters. By identifying the key deviation dimensions and adjusting the corresponding processing paths in a targeted manner, it not only effectively corrects the assessment deviation, but also maximizes the stability of the model, making the personalized assessment results more accurate and reliable.

[0069] To further improve the accuracy of parameter adjustment instruction generation, in some embodiments, step 304: generating a parameter adjustment direction instruction for the target feature extraction path based on the evaluation deviation value of the specific dimension includes: Step 401: Match the evaluation deviation value of the specific dimension with a preset clinical grading threshold set.

[0070] In step 401, the clinical grading threshold set includes physiological characteristic deviation range standards corresponding to different depression levels. These thresholds are derived from the diagnostic experience of professional medical institutions and are used to determine the clinical significance of the deviation values.

[0071] In an embodiment of the present application, the system reads a preset threshold value set, compares the deviation value of the current dimension with multiple grading standards in the set one by one, determines the severity range of the deviation value, and provides a medical basis for subsequent instruction generation.

[0072] Step 402: If the matching result indicates that the assessment deviation value is greater than the clinical moderate depression threshold in the clinical grading threshold set, a positive instruction is generated for enhancing the emotion-related feature capture capability in the target feature extraction path.

[0073] In step 402, the ability to capture emotion-related features refers to the depression assessment model's ability to identify physiological features associated with emotional states. Specifically, this is achieved by analyzing physiological indicators associated with known depressive symptoms, such as heart rate variability data, heartbeat interval fluctuation patterns, and voice intonation characteristics in speech data. This capability is derived from learning the correspondence between professional medical psychological assessment labels and multimodal physiological data during model training. Positive instructions are adjustments to strengthen the neural network's ability to recognize specific physiological features and are triggered when deviations indicate that the model is underrecognizing important emotional features.

[0074] In an embodiment of the present application, when the system finds that the deviation of a certain dimension exceeds the moderate depression threshold, it will mark the feature extraction path corresponding to the dimension and generate an enhancement instruction to increase the response sensitivity of the relevant neurons, focusing on improving the ability to capture emotion-related features.

[0075] Step 403: Alternatively, if the matching result indicates that the evaluation deviation value is less than the clinical mild depression threshold in the clinical grading threshold set, a negative instruction for suppressing noise sensitivity in the target feature extraction path is generated.

[0076] In step 403, the negative instruction refers to an adjustment command that requires reducing the sensitivity of the neural network to non-critical features, and is triggered when the deviation shows that the model has overfitting noise.

[0077] In an embodiment of the present application, for deviation values ​​below the mild depression threshold, the system will generate an inhibition instruction, requiring the appropriate reduction of the parameter weight of the corresponding path, reducing overreaction to changes in non-critical features, and improving the model's anti-interference ability.

[0078] Step 404: Dynamically set the intensity level of the positive instruction or the negative instruction according to the deviation amplitude ratio between the evaluation deviation value and the preset clinical grading threshold value to form a parameter adjustment direction instruction.

[0079] In step 404, the preset clinical grading thresholds include two critical points: the "Clinical Moderate Depression Threshold" and the "Clinical Mild Depression Threshold." The former is used to determine whether feature capture capabilities need to be enhanced, while the latter is used to determine whether noise sensitivity needs to be suppressed. Together, these two thresholds constitute a complete set of thresholds for generating instructions for different situations. The action intensity level is a quantitative indicator of adjustment force calculated based on the proportion of the deviation value exceeding the threshold, ensuring that the adjustment amplitude matches the severity of the deviation.

[0080] In an embodiment of the present application, the system calculates the difference between the deviation value and the nearest threshold value, converts it into a specific intensity level value according to a preset proportional coefficient, and finally forms a complete instruction including the adjustment direction and strength.

[0081] Here's a specific example: In user A's sixth-week monitoring data, the system detected that the deviation value in the light intensity dimension remained at +0.6 points (model prediction: 1.8 points minus the standard value: 1.2 points), while the deviation value in the voice dimension rebounded to -0.2 points (model prediction: 1.7 points minus the standard value: 1.9 points). The system first matched the light intensity deviation value with the clinical grading threshold set and confirmed that the +0.6 point exceeded the moderate depression threshold of 0.5 points but did not reach the severe threshold of 1.0 points, placing it within the range requiring enhanced treatment. The intensity level was calculated using the formula L = (ΔS - T_m) / Δu, where L represents the intensity level, ΔS represents the deviation value of 0.6 points, T_m represents the moderate threshold of 0.5 points, and Δu represents the adjustment unit of 0.25 points. The resulting intensity level is 1 (0.6-0.5) / 0.25, rounded to the nearest whole number. The voice deviation value of -0.2 points, meanwhile, just met the mild depression threshold and did not trigger any adjustment instructions. The system only generates a level 1 forward instruction for the illumination feature extraction path, increasing the weight of the neuron responsible for identifying abnormal illumination periods in this path by 0.1×1=0.1.

[0082] In the embodiments of the present application, the method uses an instruction generation mechanism guided by clinical standards to ensure that the model optimization process not only complies with medical standards but can also be flexibly adjusted according to individual differences, making the depression status assessment results more accurate and reliable while maintaining the overall stability of the model.

[0083] To further improve the accuracy and comprehensiveness of depression state assessment, in some embodiments, step 102: processing the multimodal data through the time series fusion network in the depression assessment model to output a predicted depression state assessment result includes: Step 501: The time series fusion network is used to perform parallel time series scanning on the heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data and voice data within the same period, and dynamic change features related to emotional fluctuations are extracted from each modal data.

[0084] In step 501, dynamic change features refer to specific change patterns related to emotional states identified from various types of physiological data. These features can reflect the fluctuations in the user's psychological state.

[0085] In an embodiment of the present application, the system sends the collected heart rate, skin conductance and other data into the corresponding analysis channels respectively, and detects abnormal fluctuation periods through a sliding time window, such as the continuous large fluctuation segments in the heart rate data, the segments with abnormally flat intonation in the voice data, etc., and extracts the time position and change amplitude information of these feature segments.

[0086] Step 502: establishing a dynamic response relationship between different modal data based on the time synchronization of the dynamic change characteristics.

[0087] In step 502, the temporal synchronization of dynamically changing features refers to the temporal correspondence between emotion-related feature segments identified in different physiological data sets. For example, whether periods of abnormal heart rate fluctuations coincide with periods of voice intonation changes. This temporal matching reflects the coordinated response patterns of different physiological systems to emotional changes. Dynamic response relationships refer to the temporal co-occurrence and mutual influence patterns of different physiological data features, reflecting the coordinated responses of various body systems to emotional changes. These include: the real-time correspondence between rhythmic fluctuations in heart rate variability data and changes in voice intonation in voice data; and the concomitant changes between intensity changes in ambient light data and the amplitude of wrist motion data.

[0088] In an embodiment of the present application, the system aligns the various extracted feature segments according to the time axis and analyzes the temporal correlation between them, such as whether the heart rate fluctuation increases when the voice tone becomes flatter, or whether the body movement decreases when the skin conductance increases, and establishes a correspondence table of these cross-modal features.

[0089] Step 503: Based on the dynamic response relationship and the preset professional label mapping rules, the independent contribution value of each physiological characteristic dimension to the depressive state is calculated.

[0090] In step 503, the professional label mapping rules refer to the clinical correspondence standards between various physiological characteristics and depressive symptoms provided by medical institutions, which are used to quantify the influence of these characteristics on the depressive state. Each physiological characteristic dimension specifically refers to an independent physiological indicator category related to the depressive state mapped from multimodal data, including: heart rate variability (reflecting autonomic nervous system function), skin conductance (characterizing sympathetic nerve activation), ambient light intensity (indicating circadian rhythm adaptability), wrist movement (reflecting motor behavior patterns), and speech (capturing emotional characteristics of speech). Each dimension corresponds to a specific data type's contribution analysis in depression assessment. The independent contribution value is a score of the individual impact of each physiological characteristic dimension (such as heart rate, speech, etc.) on the total depressive state score. This value is calculated by analyzing the correspondence between the dimension's characteristics and the professional medical labels. It reflects the contribution of the dimension's characteristics alone to the prediction of depressive symptoms, without considering the influence of other dimensions.

[0091] In an embodiment of the present application, the system converts the established dynamic response relationship into a specific contribution value based on preset clinical standards. For example, the combination of "heart rate fluctuation + flat tone" corresponds to a contribution value of 3 points, and the combination of "increased skin conductance + reduced body movement" corresponds to a contribution value of 2 points. The average impact value of each dimension within the cycle is taken.

[0092] Step 504: Aggregate the contribution values ​​of each physiological characteristic dimension to generate a total score of the depressive state.

[0093] In step 504 , the total score of depression status is a comprehensive evaluation result obtained by weighted summarization of the contribution values ​​of each dimension, reflecting the overall depression level of the user.

[0094] In an embodiment of the present application, the system adds the contribution values ​​of various dimensions such as heart rate and voice according to preset weights, where core indicators such as heart rate dimension have higher weights, and environmental indicators such as light intensity dimension have lower weights, and finally obtains a quantitative score reflecting the overall depression condition.

[0095] Step 505: Combining the contribution value of each physiological characteristic dimension with the total score of the depression state to form a depression state assessment result.

[0096] In an embodiment of the present application, the system packages the calculated total score and contribution values ​​of each dimension into structured data, including analysis of the main abnormal feature time periods and key influencing factors, to form a complete evaluation conclusion.

[0097] Here's a specific example: During the eighth week of monitoring for user A, the system first performed a time-series scan analysis of the collected physiological data. It discovered that between 11 PM and 1 AM, the patient exhibited a persistent heart rate fluctuation of 28ms (15ms above the baseline value). Four of the five speech segments collected during this period exhibited flattened intonation (a 42%±3% decrease in intonation). Skin conductance data also showed an increase of 3.5μS±0.5μS during this period. The system aligned these features based on timestamps, establishing a dynamic response relationship of "heart rate acceleration + flattened intonation + increased skin conductance," with 85% temporal overlap. According to professional label mapping rules, this combination corresponds to a base contribution of 3.5 points. After adjusting for time coverage, the final contribution from the heart rate dimension was 3.2 points (3.5 × 0.85 × 1.05, where 1.05 is the weighting factor for the nighttime period), the speech dimension contributed 2.9 points, and the skin conductance dimension contributed 2.1 points. Combining the daily data from the physical activity dimension (2.3 points, a 35% decrease in daily activity) and the light intensity dimension (1.8 points, less than 800 lux daily light exposure), the system aggregated the contributions from each dimension to arrive at a total depression score of 12.3 points for the week (3.2 + 2.9 + 2.1 + 2.3 + 1.8). This score deviated by 0.7 points from the user's hospital PHQ-9 score of 13 (moderate depression) for the same period. The assessment results specifically highlighted abnormal nighttime physiological indicators as the primary contributing factor, prioritizing sleep quality improvement. This conclusion was highly consistent with the doctor's diagnosis of "sleep disorders leading to emotional problems," validating the accuracy of the model analysis.

[0098] In the embodiments of the present application, the method achieves a comprehensive quantitative assessment of the depressive state through temporal correlation analysis of multi-dimensional physiological data, which can not only reflect the overall severity but also identify specific symptom manifestations, providing a reliable basis for precise intervention while ensuring the objectivity and traceability of the evaluation process.

[0099] To further improve the standardization and traceability of depression assessment results, in some embodiments, step 103: generating a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and associating the multimodal data within a preset period with the standardized depression assessment result, includes: Step 601: Arrange the predicted depression state assessment results corresponding to all periods in chronological order to form a user depression state evolution sequence.

[0100] In step 601, the user's depression state evolution sequence refers to a trend chart composed of evaluation results of each period arranged in chronological order, reflecting the overall trend of depression state changes over time and the evolution law of each dimensional feature.

[0101] In an embodiment of the present application, the system sorts the evaluation results of all cycles by week, records the changes in the weekly total score and the contribution value of each dimension, and forms a comprehensive evolution map including the time dimension, the total score curve and the characteristic curve of each dimension, which intuitively shows the development process of the user's depressive state.

[0102] Step 602: Generate a standardized depression assessment result based on the fluctuation trend of the user's depression state evolution sequence in combination with a preset clinical assessment standard.

[0103] In step 602, clinical evaluation standards refer to the depressive symptom grading rules developed by professional medical institutions. These standards clearly define the symptom manifestations and quantitative indicator ranges corresponding to different levels of depression. For example, a PHQ-9 scale score of 5-9 is defined as mild depression, and 10-14 points is defined as moderate depression. The system ensures that the evaluation results output by the model are consistent with the clinical diagnosis by presetting these authoritative medical standards as the basis for conversion.

[0104] In an embodiment of the present application, the system analyzes the evolution sequence, compares it with the preset clinical conversion criteria, and maps the continuous change trend into professional medical expressions, such as "the total score continues to decrease from 14 points to 9 points, accompanied by improvement in sleep indicators" is converted into a standardized conclusion of "mild depressive symptoms continue to improve."

[0105] Step 603: Bind the multimodal data within the complete cycle corresponding to the standardized depression assessment result one-to-one with the standardized depression assessment result according to the data collection timestamp to form an associated data set, which is the associated result.

[0106] In step 603, the associated data set refers to a matching set established by matching the original physiological monitoring data with the final evaluation conclusion in a time correspondence, ensuring that each data point can be traced back to the corresponding evaluation conclusion.

[0107] In an embodiment of the present application, the system will generate all physiological data within the period corresponding to the standardized assessment results, establish an associated index based on the collection time point and the assessment conclusion, and form a complete record set containing the original data and diagnostic conclusions for subsequent model optimization and analysis backtracking.

[0108] Here's a specific example: After User A completed eight weeks of monitoring, the system chronologically arranged their weekly depression assessment results into a progression sequence: Week 1's total score was 14 (heart rate 3.5, speech 2.8, body movement 3.2, light intensity 2.5, skin conductance 2), dropping to 9 in Week 4 (light intensity 2.8, heart rate 2.5), and further to 7.5 in Week 8 (light intensity 1.8, heart rate 2). System analysis of this sequence revealed a steady downward trend in the total score (an average weekly decrease of 0.8 points), with the light intensity dimension showing the greatest improvement (from 2.5 to 1.8). Combined with the hospital's PHQ-9 score of 7 (mild depression) in Week 8 and the doctor's comment that "sleep improved significantly," the system converted this progression into a standardized assessment result: "Sustained relief of mild depressive symptoms and improved sleep quality." The system then correlated this conclusion with the complete eighth-week data set, which included 5,760 heart rate data points (1 per minute x 24 hours x 7 days x 60% valid data), 2016 skin conductance data points (1 per 5 minutes), and 10,080 light intensity data points (1 per minute). The system precisely matched the data collection time points with the evaluation conclusions. For example, abnormal light intensity data at 2:00 AM (intensity <50 lux for 120 minutes) was correlated with the "improved sleep quality" conclusion, forming a correlated dataset encompassing all key data points from the eight weeks. This dataset showed that when the contribution of the light intensity dimension dropped below 2 points, the total score remained within 0.5 points of the hospital's diagnosis, validating the reliability of the data correlation system and providing accurate data support for subsequent personalized treatment.

[0109] In the embodiments of the present application, this method makes the model prediction results clinically interpretable by establishing a time-series evolution analysis and standardized conversion mechanism. At the same time, through a complete data association system, it not only ensures the reliability of the evaluation conclusions, but also provides complete data support for subsequent personalized optimization, thereby realizing the standardization and personalized unification of depression status assessment.

[0110] To further improve the effect of personalized optimization of the depression assessment model, in some embodiments, step 203: incrementally updating the neuron weights involved in the parameter adjustment direction instruction to generate a personalized depression assessment model adapted to the user's individual physiological response pattern includes: Step 701: Integrate the associated data set of the preset period with the associated data set accumulated in the historical optimization period, and calculate the model parameter update amount based on the incremental learning method.

[0111] In step 701, the integration of linked datasets involves merging and analyzing newly generated data with historical optimization data, retaining validated features while incorporating newly discovered pattern changes. The model parameter update refers to the neural network weight adjustment scheme calculated using an incremental learning algorithm. Specifically, it includes the location of the parameter to be modified, the direction of adjustment, and the magnitude of the change. These quantitative indicators clearly define which neuronal connections in the model need to be strengthened or weakened, as well as the specific magnitude of the adjustment, ensuring that the model optimization process is both accurate and controllable.

[0112] In an embodiment of the present application, the system time-aligns and feature-matches the data collected in the current cycle with all previously stored cycle data, identifies recurring physiological patterns and newly emerging abnormal features, calculates the parameter range and amplitude that need to be adjusted, and forms a complete update plan.

[0113] Step 702: Iteratively optimize the unified depression assessment model based on the model parameter update amount to generate a personalized depression assessment model, and apply the personalized depression assessment model to the multimodal data analysis of the next cycle to obtain a depression assessment result.

[0114] In step 702, iterative optimization refers to making targeted improvements to specific parts while maintaining the core functions of the model, ensuring that the model can adapt to new changes while maintaining stability.

[0115] In an embodiment of the present application, the system locks the specific processing paths that need to be adjusted in the model based on the calculated update plan, fine-tunes the neuron connection weights in these paths, and freezes the parameters of other irrelevant paths, generating an optimized model exclusively for the current user and immediately putting it into use in the next cycle.

[0116] Step 703: Repeat the above steps to gradually adapt the personalized depression assessment model to the user's unique physiological behavior, thereby achieving a gradual improvement in assessment accuracy.

[0117] In step 703, progressive improvement refers to making the model gradually and accurately adapt to the user's unique physiological characteristic pattern through multiple small adjustments.

[0118] In the embodiment of the present application, the system repeats the data integration and model optimization process after each monitoring cycle, so that the model's ability to identify user characteristics is continuously enhanced, and the consistency between the evaluation results and the actual situation is continuously improved.

[0119] Here's a specific example: After the 12th week of monitoring for user A, the system integrated the latest data with the associated datasets from the previous 11 weeks for analysis. The system found that the deviation in the light intensity dimension had stabilized at +0.3 points (the difference between the model's prediction of 1.8 and the standard value of 1.5 points), but the newly emerged deviation in the voice dimension remained at -0.4 points (the difference between the model's prediction of 1.6 and the standard value of 2.0 points). The system calculated the deviations in each dimension using the formula ΔS = S_model - S_clinic. The deviation in the voice dimension exceeded the mild depression threshold of -0.2 points. The system calculated the intensity level using the formula (ΔS - T) / Δu, where T is the threshold of -0.2 points and Δu is the adjustment unit of 0.15 points). This determined that the voice path required two levels of adjustment (-0.4 + 0.2) / 0.15, rounded to the nearest whole number. During the incremental optimization process, the system maintained stable parameters like the illumination path, adjusting only the weights of two groups of neurons in the speech emotion analysis path responsible for intonation recognition. The original weights W_j were updated to W_j' = W_j - β·M (where β = 0.05 is the learning rate for the speech path, and M = 2 is the intensity level). In the 13th week evaluation, the optimized model's deviation in the speech dimension narrowed to -0.1 points (from a predicted value of 1.9 points), while the deviation in the light intensity dimension remained at +0.3 points. The predicted total score deviated from the hospital's PHQ-9 score of 7.5 by 0.2 points. Through 12 weeks of incremental optimization, the model accurately grasped the physiological characteristic pattern of User A's "light sensitivity + intonation suppression," improving the consistency of the evaluation results with the clinical diagnosis.

[0120] In the embodiments of the present application, this method enables the depression assessment model to dynamically track the user's physiological changes through continuous data accumulation and precise incremental optimization, thereby maintaining the stability of the core assessment function and timely adapting to the evolution of individual characteristics, thereby achieving truly personalized and accurate assessment.

[0121] Figure 2 This is a structural diagram of a depression assessment titration optimization system based on professional tags provided in an embodiment of the present application, such as Figure 2 As shown, the system includes: The acquisition module 21 is used to collect multimodal data of the user according to a preset period, and the multimodal data includes: heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data and voice data.

[0122] The input module 22 is used to input the multimodal data within the same period into a unified depression assessment model trained with professional labels, process the multimodal data through the time series fusion network in the depression assessment model, and output a predicted depression state assessment result, which includes the total depression state score and the contribution value of each physiological characteristic dimension.

[0123] The generating module 23 is configured to generate a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and associate the multimodal data within a preset period with the standardized depression assessment result.

[0124] The optimization module 24 is configured to optimize the parameters of the unified depression assessment model through incremental learning based on the correlation results, so as to generate a personalized depression assessment model for the user and realize progressive titration optimization of depression assessment.

[0125] Figure 2 The depression assessment titration optimization system based on professional labels can be performed Figure 1 The implementation principles and technical effects of the professional tag-based depression assessment titration optimization method described in the illustrated embodiment are not further elaborated. The specific manner in which each module and unit performs operations in the professional tag-based depression assessment titration optimization system described in the above embodiment has been described in detail in the related embodiments and will not be further elaborated here.

[0126] In one possible design, Figure 2 The depression assessment titration optimization system based on professional labels of the embodiment shown can be implemented as a computing device, such as Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are called and executed by the processing component 32 .

[0127] The processing component 32 is used to perform the above Figure 1 The embodiment provides a depression assessment titration optimization method based on professional labels.

[0128] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above method. Of course, the processing component may also be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above method.

[0129] The storage component 31 is configured to store various types of data to support operations on the terminal. The storage component can be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random-access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk, or optical disk.

[0130] Of course, a computing device may also include other components, such as input / output interfaces, display components, communication components, etc.

[0131] The input / output interface provides an interface between the processing component and the peripheral interface module, which can be an output device, an input device, etc.

[0132] The communication component is configured to facilitate, among other things, wired or wireless communications between the computing device and other devices.

[0133] Among them, the computing device can be a physical device or an elastic computing host provided by a cloud computing platform, etc. In this case, the computing device can refer to a cloud server, and the above-mentioned processing components, storage components, etc. can be basic server resources rented or purchased from the cloud computing platform.

[0134] The present application also provides a computer storage medium storing a computer program, wherein the computer program can achieve the above-mentioned Figure 1 The illustrated embodiment is a depression assessment titration optimization method based on professional labels.

[0135] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0136] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0137] Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented using software plus a necessary general-purpose hardware platform, or of course, hardware. Based on this understanding, the essence of the above technical solution, or the portion that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, or an optical disk, and includes a number of instructions for causing a computer device (such as a personal computer, server, or network device) to execute the methods described in each embodiment or certain portions of the embodiments.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A depression assessment titration optimization method based on professional labels, characterized in that: include: Collecting multimodal data of the user according to a preset period, wherein the multimodal data includes: heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data, and voice data; Inputting the multimodal data within the same period into a unified depression assessment model trained with professional labels, processing the multimodal data through a temporal fusion network in the depression assessment model, and outputting a predicted depression state assessment result, the depression state assessment result including a total depression state score and a contribution value of each physiological characteristic dimension; Generating a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and associating the multimodal data within a preset period with the standardized depression assessment result; Based on the association results, the parameters of the unified depression assessment model are optimized through incremental learning to generate a personalized depression assessment model for the user, thereby achieving progressive titration optimization of depression assessment.

2. The method according to claim 1, characterized in that The method of optimizing the parameters of the unified depression assessment model by incremental learning based on the association results to generate a personalized depression assessment model for the user includes: Compare the standardized depression assessment results in the correlation results with the predicted depression state assessment results, and calculate the assessment deviation value of each physiological characteristic dimension based on the comparison results; Determining a parameter adjustment direction instruction for a corresponding feature extraction path in the time series fusion network according to the evaluation deviation value; Through incremental learning, while retaining the original parameter structure of the unified depression assessment model, the neuron weights involved in the parameter adjustment direction instructions are incrementally updated to generate a personalized depression assessment model that adapts to the user's individual physiological response pattern.

3. The method according to claim 2, characterized in that The step of determining, based on the evaluation deviation value, a parameter adjustment direction instruction for a corresponding feature extraction path in the time series fusion network includes: Analyzing the distribution of the evaluation deviation values ​​in each physiological characteristic dimension; Determining, based on the distribution, a specific dimension whose deviation contribution value exceeds a preset deviation contribution threshold; According to the correspondence between each physiological feature dimension and the feature extraction path in the time series fusion network, determining the neural network processing path associated with the specific dimension as the target feature extraction path; Based on the size of the evaluation deviation value of the specific dimension, a parameter adjustment direction instruction for the target feature extraction path is generated.

4. The method according to claim 3, characterized in that The generating of a parameter adjustment direction instruction for the target feature extraction path based on the evaluation deviation value of the specific dimension includes: Matching the assessment deviation value of the specific dimension with a preset clinical grading threshold set; If the matching result indicates that the assessment deviation value is greater than the clinical moderate depression threshold in the clinical grading threshold set, generating a positive instruction for enhancing the emotion-related feature capture capability in the target feature extraction path; Alternatively, if the matching result indicates that the assessment deviation value is less than the clinical mild depression threshold value in the clinical grading threshold set, generating a negative instruction for suppressing noise sensitivity in the target feature extraction path; According to the deviation amplitude ratio between the evaluation deviation value and the preset clinical grading threshold, the intensity level of the positive instruction or the negative instruction is dynamically set to form a parameter adjustment direction instruction.

5. The method according to claim 1, characterized in that The processing of the multimodal data by the time series fusion network in the depression assessment model to output a predicted depression state assessment result includes: The time series fusion network performs parallel time series scanning on the heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data, and voice data within the same period, and extracts dynamic change features related to emotional fluctuations from each modal data; establishing a dynamic response relationship between different modal data based on the time synchronization of the dynamic change characteristics; Based on the dynamic response relationship and the preset professional label mapping rules, calculating the independent contribution value of each physiological characteristic dimension to the depressive state; Aggregate the contribution values ​​of each physiological characteristic dimension to generate a total score of depression state; The contribution value of each physiological characteristic dimension and the total score of the depression state are combined to form a depression state assessment result.

6. The method according to claim 1, characterized in that Generating a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and associating the multimodal data within a preset period with the standardized depression assessment result, includes: Arrange the predicted depression status assessment results corresponding to all periods in chronological order to form a user depression status evolution sequence; Generate a standardized depression assessment result based on the fluctuation trend of the user's depression state evolution sequence and in combination with preset clinical assessment criteria; The multimodal data within a complete cycle corresponding to the standardized depression assessment result is bound one-to-one with the standardized depression assessment result according to the data collection timestamp to form an associated data set, which is the associated result.

7. The method according to claim 2, characterized in that The incremental updating of the neuron weights involved in the parameter adjustment direction instruction to generate a personalized depression assessment model adapted to the user's individual physiological response pattern includes: Integrate the associated dataset of the preset period with the associated dataset accumulated during the historical optimization period, and calculate the model parameter update amount based on the incremental learning method; Iteratively optimizing the unified depression assessment model according to the model parameter update amount to generate a personalized depression assessment model, and applying the personalized depression assessment model to the multimodal data analysis of the next cycle to obtain a depression assessment result; The above steps are repeated to gradually adapt the personalized depression assessment model to the user's unique physiological behavior, thereby achieving a gradual improvement in assessment accuracy.

8. A depression assessment titration optimization system based on professional labels, characterized by: include: An acquisition module is used to collect multimodal data of the user according to a preset period, wherein the multimodal data includes: heart rate variability data, skin conductance data, ambient light intensity data, wrist body movement data, and voice data; An input module is configured to input the multimodal data within the same period into a unified depression assessment model trained with professional labels, process the multimodal data through a temporal fusion network in the depression assessment model, and output a predicted depression state assessment result, wherein the depression state assessment result includes a total depression state score and a contribution value of each physiological characteristic dimension; A generation module, configured to generate a standardized depression assessment result for the user based on the predicted depression state assessment results corresponding to all periods, and to associate the multimodal data within a preset period with the standardized depression assessment result; The optimization module is used to optimize the parameters of the unified depression assessment model through incremental learning based on the association results to generate a personalized depression assessment model for the user and realize progressive titration optimization of depression assessment.

9. A computing device, characterized in that The method comprises a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are used to be called and executed by the processing component to implement a depression assessment titration optimization method based on professional labels as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that A computer program is stored, and when the computer program is executed by a computer, the method for optimizing depression assessment titration based on professional labels according to any one of claims 1 to 7 is implemented.

Images