A method and system for processing migraine symptom data

Abstract

The application discloses a migraine symptom data processing method and system, relates to the field of migraine symptom data processing, and is used for optimizing the consumption of computing resources and ensuring the timeliness and accuracy of migraine attack early warning and comprises the following steps: receiving data streams from physiological sensors and environmental detectors, and performing preliminary fusion processing on the data streams to obtain preliminary fusion data; inputting the preliminary fusion data into a first-level analysis module, evaluating the preliminary fusion data by the first-level analysis module, and obtaining a prediction result and a prediction confidence; when the prediction confidence is greater than a first threshold, maintaining current computing resources and outputting the prediction result; when the prediction confidence is less than or equal to the first threshold and greater than or equal to a second threshold, additionally allocating computing resources to call a second-level analysis module; and the second-level analysis module extracts features and recognizes patterns from the preliminary fusion data by using the allocated computing resources to obtain the prediction result and cause insight.

Description

Technical Field

[0001] This invention relates to the field of migraine symptom data processing, and more particularly to a method and system for processing migraine symptom data. Background Technology

[0002] In the field of healthcare, people increasingly hope to manage their health through intelligent means, especially for chronic and recurring diseases such as migraines.

[0003] For example, there is a system that aims to revolutionize the way migraine symptoms are managed, which uses wearable sensors to acquire the user's brain signals, autonomic nervous system responses such as heart rate, skin pressure levels, light intensity, sound levels, and air quality parameters.

[0004] All this raw data is transmitted in real time to a remote cloud computing platform. This platform can then identify subtle physiological changes or environmental triggers that may precede a migraine attack from this data.

[0005] However, excessive data volume can increase the platform's load. Furthermore, to extract meaningful insights from this seemingly unrelated data, these different data streams must be precisely synchronized and integrated. Simple timestamps alone are often insufficient; when attempting to identify complex migraine aura patterns, even minute temporal misalignments between different physiological signals can lead to incorrect judgments.

[0006] In migraine symptom data processing methods, how to efficiently extract complex features and perform deep inference with limited cloud computing resources to ensure the generation of highly accurate and interpretable migraine attack warnings and trigger insights in real time is an urgent technical problem to be solved. Summary of the Invention

[0007] This invention provides a method for processing migraine symptom data, which optimizes the consumption of computing resources on a cloud computing platform while ensuring the timeliness and accuracy of migraine attack warnings.

[0008] In a first aspect, to address the aforementioned technical problems, this invention provides a method for processing migraine symptom data, comprising: receiving data streams from physiological sensors and environmental detectors, and performing preliminary fusion processing on the data streams to obtain preliminary fused data; inputting the preliminary fused data into a first-level analysis module, which evaluates the preliminary fused data to obtain a prediction result and a prediction confidence level; when the prediction confidence level is greater than a first threshold, maintaining current computing resources and outputting the prediction result; when the prediction confidence level is less than or equal to the first threshold and greater than or equal to a second threshold, allocating additional computing resources to invoke a second-level analysis module; the second-level analysis module using the allocated computing resources to perform feature extraction and pattern recognition on the preliminary fused data to obtain a prediction result and a trigger insight, wherein the trigger insight includes explanatory information for the prediction result.

[0009] Optionally, when the prediction confidence is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to invoke the second-level analysis module, including: monitoring the trend of prediction confidence within a preset time window and checking the fluctuation amplitude of underlying lightweight physiological characteristics in the preliminary fusion data within the preset time window; determining whether the trend of prediction confidence is lower than a preset trend threshold and whether the fluctuation amplitude of underlying lightweight physiological characteristics is lower than a preset fluctuation threshold; when the trend of prediction confidence is lower than the preset trend threshold and the fluctuation amplitude of underlying lightweight physiological characteristics is lower than the preset fluctuation threshold, initiating the pre-deep analysis process; the pre-deep analysis process identifies non-physiological artifact patterns in the preliminary fusion data; when the pre-deep analysis process confirms the existence of real physiological abnormalities, additional computing resources are allocated to invoke the second-level analysis module.

[0010] Optionally, the monitoring of the trend of prediction confidence within a preset time window and the examination of the fluctuation range of the underlying lightweight physiological characteristics in the preliminary fusion data within the preset time window include: continuously tracking the individual physiological baseline; calculating the instantaneous value of the underlying lightweight physiological characteristics in the preliminary fusion data; comparing the instantaneous value with the individual physiological baseline to obtain the relative deviation value; adjusting the preset trend threshold and preset fluctuation threshold based on the relative deviation value and the instantaneous value of prediction confidence; and using the adjusted preset trend threshold and preset fluctuation threshold to determine the trend of prediction confidence and the fluctuation range of the underlying lightweight physiological characteristics.

[0011] Optionally, when the trend of the predicted confidence level is lower than a preset trend threshold and the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than a preset fluctuation threshold, a pre-deep analysis process is initiated, including: extracting multi-dimensional features from the preliminary fused data to obtain multi-dimensional features; checking whether the multi-dimensional features are accompanied by synchronous changes in other physiological modalities to obtain the synchronous change situation; adjusting the judgment priority of artifact recognition based on the synchronous change situation; maintaining a dynamically updated physiological pattern feature library; comparing the identified patterns with the physiological pattern feature library to obtain the comparison results; and determining the pattern type based on the comparison results.

[0012] Optionally, maintaining a dynamically updated physiological pattern feature library includes: when a new pattern is identified, performing a preliminary physiological relevance assessment based on the new pattern's duration, intensity change rate, and synchronicity with other physiological modalities to obtain preliminary assessment results; based on the preliminary assessment results, checking whether the new pattern is similar to the characteristic patterns of known non-migraine physiological events to obtain similarity check results; when the similarity check results show that the new pattern has a high similarity to the characteristic patterns of non-migraine physiological events greater than or greater than a similarity threshold, marking the new pattern as a non-migraine-related pattern and storing it in a non-migraine pattern temporary storage area; when the similarity check results show that the new pattern has a similarity to the characteristic patterns of non-migraine physiological events less than a similarity threshold, initiating a multi-round expert verification process to independently evaluate the new pattern and obtain expert verification results; and based on the expert verification results, incorporating the new pattern into the dynamically updated physiological pattern feature library.

[0013] Optionally, a multi-round expert verification process can be initiated to independently evaluate the new model and obtain expert verification results, including: distributing the new model to multiple experts for independent evaluation and obtaining the evaluation results of each expert; when there are inconsistencies in the evaluation results of the experts, obtaining expert verification results based on expert discussion meetings.

[0014] Optionally, based on expert discussion meetings, expert validation results are obtained, including: initiating expert discussion meetings; presenting multi-dimensional data analysis results of the new model, expert evaluation records of historical similar models, and relevant clinical guidelines during the expert discussion meetings; wherein, the multi-dimensional data analysis results include signal spectral characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics; the expert evaluation records include evaluation time, evaluation experts, evaluation conclusions, and evaluation basis; the relevant clinical guidelines include migraine diagnostic criteria, key points for differential diagnosis, and treatment intervention recommendations; based on the multi-dimensional data analysis results of the new model, the expert evaluation records of historical similar models, and relevant clinical guidelines, experts are assisted in reaching a consensus on the evaluation results, thus obtaining expert validation results.

[0015] Optionally, the multi-dimensional data analysis results of the new model can be displayed, including: constructing an interactive data visualization interface that displays the multi-dimensional data analysis results of the new model in real time. The multi-dimensional data analysis results include signal spectral characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics. The interactive data visualization interface allows experts to dynamically zoom, filter, and cross-compare the signal spectral characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics. The interactive data visualization interface automatically highlights or provides specific dimension interpretation assistance information based on the expert's preset focus or professional field.

[0016] Optionally, the interactive data visualization interface automatically highlights or provides specific dimensions of interpretive aids based on the expert's preset focus or area of ​​expertise. This includes: the interactive data visualization interface identifying the expert's preset focus or area of ​​expertise; the interactive data visualization interface assessing the degree of matching between the focus or area of ​​expertise and the actual physiological performance of the new pattern; when the degree of matching is lower than a preset matching threshold, the interactive data visualization interface prompting the expert that there is a potential bias; and based on the potential bias, the interactive data visualization interface recommending other relevant dimensions of interpretive aids or providing cross-domain interpretation guidance to guide the expert in understanding the new pattern.

[0017] Secondly, the present invention provides a migraine symptom data processing system, the system comprising:

[0018] The data receiving and fusion module is used to receive data streams from physiological sensors and environmental detectors, and to perform preliminary fusion processing on the data streams to obtain preliminary fused data;

[0019] The first-level analysis module is used to input the preliminary fused data into the first-level analysis module. The first-level analysis module evaluates the preliminary fused data and obtains the prediction results and prediction confidence.

[0020] The resource allocation module is used to maintain the current computing resources and output the prediction results when the prediction confidence is greater than the first threshold.

[0021] When the prediction confidence is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to call the second-level analysis module;

[0022] The second-level analysis module is used to extract features and recognize patterns from the preliminarily fused data using the allocated computing resources, so as to obtain prediction results and causal insights. Causal insights include explanatory information for the prediction results.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] First, addressing the challenges of large data volumes and high real-time requirements, this application employs a two-level analysis module design to achieve dynamic allocation of computing resources. When the prediction confidence is high, only the first-level analysis module is used, avoiding unnecessary deep computations, thereby reducing the computational burden and improving processing efficiency and real-time response capabilities.

[0025] Secondly, addressing the issue of difficulty in providing interpretable insights, this application, when the prediction confidence is low, invokes a second-level analysis module for deeper feature extraction and pattern recognition, generating causal insights containing explanatory information. This enables patients and clinicians to understand the physiological or environmental factors behind the prediction results, thereby achieving more precise intervention and management, overcoming the lack of interpretability in traditional "black box" models.

[0026] Finally, addressing the issue of low resource allocation efficiency, this application intelligently adjusts the allocation of computing resources based on prediction confidence, avoiding resource waste and increased operating costs caused by performing high-intensity computations for all data streams. This on-demand allocation strategy significantly optimizes the resource utilization efficiency of the cloud computing platform while ensuring prediction accuracy, achieving a balance between low latency and high accuracy.

[0027] In summary, this application, through its hierarchical analysis and dynamic resource allocation strategy, not only improves the efficiency and accuracy of migraine symptom data processing but also provides clinically meaningful interpretable insights. It effectively addresses multiple challenges in data processing, resource management, and interpretability in existing technologies, providing a superior technical solution for early warning and personalized management of migraines. Attached Figure Description

[0028] Figure 1 This is a schematic flowchart of a migraine symptom data processing method provided in an embodiment of the present invention;

[0029] Figure 2 This is a schematic diagram of another migraine symptom data processing method provided in an embodiment of the present invention;

[0030] Figure 3 This is a schematic diagram of the structure of a migraine symptom data processing system provided in an embodiment of the present invention. Detailed Implementation

[0031] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0032] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0033] To better understand the migraine symptom data processing method proposed in this application, the key terms and implementation environment involved will be explained in detail below.

[0034] Physiological sensors are devices used to monitor human physiological parameters in real time, such as heart rate sensors, electroencephalogram (EEG) sensors, and skin conductance sensors. These sensors can capture physiological signals associated with migraine attacks, providing a basis for subsequent data analysis.

[0035] Environmental detectors are devices used to monitor parameters of the surrounding environment, such as light sensors, temperature sensors, humidity sensors, and air quality sensors. These detectors can acquire environmental factors that may trigger migraines, providing support for comprehensive analysis.

[0036] Data stream: refers to the real-time data sequence continuously generated and transmitted by physiological sensors and environmental detectors. This data is generated continuously and at a high frequency, requiring efficient receiving and processing mechanisms.

[0037] Trigger Insight: This refers to the explanatory information provided by the second-level analysis module along with the prediction results, such as which physiological or environmental factors significantly influence the likelihood of migraine attacks. This helps patients and doctors understand the reasons behind the prediction.

[0038] The following specific embodiments will provide a detailed description and explanation of a migraine symptom data processing method provided in this application.

[0039] Reference Figure 1 This invention provides a method for processing migraine symptom data, comprising the following steps:

[0040] S1 receives data streams from physiological sensors and environmental detectors, and performs preliminary fusion processing on the data streams to obtain preliminary fused data.

[0041] Preliminary fusion data refers to the data obtained after initial integration and preprocessing of data streams from different physiological sensors and environmental detectors. Preliminary fusion processing may include data cleaning, noise reduction, time synchronization, and format standardization, aiming to provide high-quality input for subsequent analysis modules.

[0042] As one possible implementation, the system can use timestamp alignment to synchronize data from different sources, ensuring that all data points correspond to the same timeline. The raw data can then undergo preliminary cleaning and noise reduction to obtain preliminary fused data.

[0043] For example, the system can remove transient noise using a moving average filter or a median filter, or remove outliers using a simple threshold. For instance, transient spikes in heart rate data may be identified as artifacts and smoothed out.

[0044] S2. Input the preliminary fused data into the first-level analysis module. The first-level analysis module evaluates the preliminary fused data and obtains the prediction results and prediction confidence.

[0045] The first-level analysis module is a lightweight and efficient analysis unit whose main function is to quickly evaluate the preliminary fused data to generate preliminary prediction results and corresponding prediction confidence levels. This module typically uses machine learning models or heuristic rules with low computational complexity.

[0046] As one possible implementation, the system can determine the migraine risk prediction result as the occurrence of migraine when a preset proportion of preset indicators in the preliminary fused data is outside the indicator threshold, and determine the number of preset indicators outside the indicator threshold as the corresponding prediction confidence level.

[0047] The preset ratio can be set as needed. For example, it can be set to 50%.

[0048] S3. When the prediction confidence is greater than the first threshold, maintain the current computing resources and output the prediction result.

[0049] The first threshold can be set as needed. For example, it can be 90%.

[0050] Prediction results can be sent directly to users or healthcare providers via a user interface (such as a mobile app) or a notification system (such as SMS or email). For example, if the first-level analysis module predicts that a user has a high probability of experiencing a migraine in the next few hours, and the confidence level reaches 90% (above the first threshold), the system will immediately send an alert to the user and suggest preventative measures without initiating deeper analysis.

[0051] S4. When the prediction confidence is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to call the second-level analysis module.

[0052] These resources may include more processor cores, more memory, or a faster graphics processing unit (GPU) to support more complex calculations in the second-level analysis module.

[0053] The second threshold can be set as needed. For example, it can be 30%.

[0054] For example, if the prediction confidence of the first-level analysis module is 60% (between the first and second thresholds), the system will determine that more detailed analysis is needed and request more computing resources from the server cluster where the second-level analysis module is located.

[0055] As one possible implementation, the system can allocate additional computing resources based on the following steps to invoke the second-level analysis module:

[0056] S41. Monitor the trend of prediction confidence within a preset time window, and check the fluctuation range of underlying lightweight physiological characteristics in the preliminary fusion data within the preset time window.

[0057] The preset time window can be configured according to the actual application scenario and data characteristics, such as being set to 5 seconds, 10 seconds or longer, to capture short-term or medium-term confidence dynamics. The underlying lightweight physiological characteristics can include heart rate variability, skin conductance response, respiratory rate, etc.

[0058] As one possible implementation, the system can calculate the average rate of change, standard deviation, or trend slope over the past few seconds or minutes as the trend of the prediction confidence level within a preset time window. For the same underlying lightweight physiological feature, the difference between the maximum and minimum values ​​of the underlying lightweight physiological feature within the preset time window is determined as the fluctuation amplitude.

[0059] S42. Determine whether the trend of the predicted confidence level change is lower than the preset trend threshold, and whether the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than the preset fluctuation threshold.

[0060] Among them, the preset trend threshold and preset fluctuation threshold can be set as needed and are not restricted.

[0061] S43. When the trend of the predicted confidence level is lower than the preset trend threshold and the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than the preset fluctuation threshold, start the pre-deep analysis process.

[0062] Among them, the pre-depth analysis process identifies non-physiological artifact patterns in the preliminary fused data.

[0063] Non-physiological artifact patterns can be understood as data anomalies caused by non-biological factors such as sensor malfunction, improper wear, environmental interference (e.g., electromagnetic interference, motion artifacts), or data transmission errors. Pre-deep analysis can employ lightweight machine learning models or rule-based algorithms to perform rapid pattern matching and anomaly detection on the data to distinguish these artifacts from genuine physiological signals. For example, the presence of artifacts can be determined by analyzing the signal's spectral characteristics, instantaneous amplitude anomalies, or the absence of specific frequency components.

[0064] The process of identifying non-physiological artifact patterns involves the user wearing wearable devices, including ECG sensors, skin conductance sensors, and a triaxial accelerometer.

[0065] Acquire electrocardiogram (ECG) signals and triaxial accelerometer (TAI) signals; compare the ECG and TAI signals with the ECG and TAI signals corresponding to known "motion artifacts" patterns in the physiological pattern feature library; if the similarity between the two is greater than the similarity threshold, it is determined to be a non-physiological artifact pattern.

[0066] Alternatively, if the skin conductance (EDA) signal suddenly drops to near zero and persists for a preset duration, it can be identified as a non-physiological artifact pattern.

[0067] S44. When the pre-deep analysis process confirms the presence of real physiological abnormalities, additional computing resources are allocated to call the second-level analysis module.

[0068] Among these, true signs of physiological abnormality may include, but are not limited to: weak but persistent patterns in the data that are associated with known physiological events (such as heartbeat and breathing) despite low confidence, or data that show specific biomarker patterns associated with migraine aura or onset after excluding artifacts.

[0069] Understandably, through the above technical solutions, this application can significantly improve the decision-making accuracy and resource utilization efficiency of the migraine symptom data processing system under uncertain conditions. Specifically, by comprehensively judging the trend of prediction confidence changes and the fluctuation amplitude of underlying lightweight physiological characteristics, the system can more intelligently identify the true physiological complexity that requires in-depth analysis, rather than simply responding to any low-confidence situation. In addition, the introduction of the pre-deep analysis process enables the system to effectively identify and exclude non-physiological artifact patterns in the preliminary fusion data before calling the second-level analysis module, thereby avoiding time-consuming and potentially misleading in-depth analysis of noisy or erroneous data. As a result, not only is unnecessary consumption of computing resources reduced and the system's operating costs lowered, but the second-level analysis module can also focus on processing high-quality data with real physiological significance, thereby improving the reliability and accuracy of the final prediction results and trigger insights, and providing users with more accurate migraine management recommendations.

[0070] In some preferred embodiments, a specific example is given below. Suppose a user is wearing a physiological sensor, and the system is running a migraine symptom data processing method. At a certain moment, the prediction confidence output by the first-level analysis module decreases, falling between a first threshold and a second threshold, indicating uncertainty in the current situation.

[0071] According to the scheme of this application, the system will first monitor the trend of the prediction confidence over the past 30 seconds, and check the fluctuation of the underlying lightweight physiological characteristics (such as heart rate and skin conductance) in the preliminary fusion data within the same 30-second time window.

[0072] For example, if the monitoring results show that although the prediction confidence is low, its trend is very flat within the time window, without a sharp decline or abnormal fluctuation; at the same time, the fluctuation amplitude of the underlying lightweight physiological characteristics is also lower than the preset fluctuation threshold, showing abnormal stability, this may indicate that there is some non-physiological interference in the data, rather than a real physiological change.

[0073] In this situation, the system will initiate a pre-deep analysis process. This process may analyze the spectral characteristics of the preliminary fused data and discover periodic noise at specific frequencies, or detect sudden interruptions or saturation of sensor signals, all of which are identified as non-physiological artifact patterns.

[0074] Since the pre-deep analysis process confirms the presence of non-physiological artifacts rather than genuine physiological abnormalities, the system will not allocate additional computing resources to invoke the second-level analysis module. Instead, the system may attempt data cleaning or prompt the user to check the sensor wearing status.

[0075] Conversely, if the predicted confidence trend is lower than the preset trend threshold, but the fluctuation of the underlying lightweight physiological characteristics shows a subtle but persistent pattern related to known physiological events (such as mild arrhythmia or changes in breathing patterns), and the pre-deep analysis process eliminates artifacts and confirms the existence of real physiological abnormalities, then the system will allocate additional computing resources to call the second-level analysis module for more in-depth feature extraction and pattern recognition to reveal potential migraine triggers.

[0076] In this way, the proposed solution avoids unnecessary in-depth analysis when data quality is questionable or artifacts exist, thereby improving resource utilization efficiency and analysis accuracy.

[0077] S5, the second-level analysis module, utilizes allocated computing resources to perform feature extraction and pattern recognition on the preliminary fused data to obtain prediction results and insights into the underlying causes.

[0078] Among them, the trigger insight includes explanatory information for the prediction results.

[0079] As one possible implementation, the system can employ deep learning techniques, such as convolutional neural networks (CNNs) or recurrent neural networks (RNNs), to automatically learn and extract high-dimensional, abstract physiological and environmental features from the initially fused data. These features may include complex patterns of heart rate variability, changes in specific frequency components of brain waves, and the combined effects of environmental factors. For pattern recognition, more sophisticated machine learning models, such as long short-term memory networks (LSTMs) or Transformer models, can be used to deeply analyze the extracted features to identify complex patterns highly correlated with migraine attacks, thereby obtaining more accurate predictions. Simultaneously, to provide insights into the triggers, the second-level analysis module can use methods such as LIME (Local Interpretable Model-agnostic Explanations) or SHAP (SHapley Additive explanations) to analyze which input features contribute most to the prediction results when the model makes predictions, thus generating explanatory information for the predictions.

[0080] For example, the system might indicate that "your heart rate variability has decreased significantly in the past two hours, while the ambient light intensity has remained too high, which may be a major trigger for a migraine attack."

[0081] The following explains the generation mechanism of the explanatory information for the prediction results contained in the causal insight:

[0082] 1. Feature Extraction and Pattern Recognition: The second-level analysis module first performs in-depth feature extraction and pattern recognition on the initially fused data. This may involve complex deep learning models (such as Convolutional Neural Networks (CNNs) or Recurrent Neural Networks (RNNs), which can automatically learn and identify high-dimensional, abstract physiological and environmental feature patterns that are highly correlated with migraine attacks from data from physiological sensors (such as EEG signals, heart rate variability, and skin conductance) and environmental detectors (such as light intensity, sound levels, and air quality).

[0083] 2. Integration of Explainable Artificial Intelligence (XAI) Technology: After obtaining the prediction results, the second-level analysis module calls the integrated XAI algorithms. These algorithms do not directly change the prediction model itself, but rather quantify the influence of each input feature on the final prediction result by analyzing the input-output relationship of the model or the specific structure inside the model.

[0084] LIME works by locally perturbing the raw data (e.g., making minor modifications to certain physiological or environmental parameters in the initial fused data) and then observing how these perturbations affect the predictions of the second-level analysis module. By constructing a simple, interpretable local model (such as a linear model) to approximate the behavior of complex models, LIME can identify which feature changes have the greatest impact on predictions in a specific prediction context. SHAP, based on cooperative game theory, assigns a Shapley value to each feature, representing its contribution to the model's predictions. It provides a globally consistent and locally accurate interpretation by fairly allocating the contribution of each feature by considering all possible combinations. This means that SHAP can quantify the specific impact of each physiological or environmental feature (e.g., decreased heart rate variability, increased light intensity) on the probability of increasing or decreasing migraine predictions.

[0085] 3. Generating Explanatory Information: After the XAI technology analysis is completed, it outputs the contribution of each feature to the prediction result. The second-level analysis module transforms these features with the highest contribution and their direction of influence into human-readable and clinically meaningful explanatory information as part of the causal insight. This explanatory information aims to clarify "why a certain prediction was made" and "what specific physiological or environmental factors are contributing to potential migraine attacks".

[0086] For example, suppose the second-level analysis module predicts that a user has a high probability of experiencing a migraine in the next few hours. Through integrated XAI technology analysis, the system might identify the following features that contribute most to this prediction:

[0087] Heart rate variability (HRV): A significant decrease in the past two hours, which is usually associated with autonomic nervous system imbalance or increased stress.

[0088] Ambient light intensity: If it has remained at a high level for the past hour, this can be a common trigger for migraines.

[0089] Sleep duration: Sleep duration below average in the past 24 hours. Sleep deprivation is another common trigger for migraines.

[0090] Based on these analyses, the second-level analysis module will generate the following explanatory information as part of the trigger insight: "Your heart rate variability has decreased significantly in the past two hours, while the ambient light intensity has remained too high, and you have not had enough sleep in the past 24 hours. These factors together increase the likelihood of a migraine attack." Such explanatory information can help users understand the specific reasons behind the prediction, thereby enabling them to take targeted interventions, such as rest, avoiding bright light, or engaging in relaxation training.

[0091] Thus, compared to the closest existing technologies, the advantages of this application lie in its intelligent management of computing resources and the interpretability of prediction results. While existing technologies can also perform multimodal data analysis and prediction, they often have the following limitations: first, the lack of a refined resource allocation mechanism may lead to the investment of large amounts of computing resources in all situations, resulting in resource waste and increased costs; second, their prediction models are usually "black boxes," making it difficult to provide explanations for the prediction results, making it difficult for patients and doctors to understand the reasons for the predictions, thereby limiting the effectiveness of interventions. This application achieves a balance between computational efficiency and prediction accuracy by introducing a two-level analysis module and dynamic resource allocation based on confidence levels. More importantly, the causal insights provided by the second-level analysis module greatly enhance the system's usability and user trust, enabling migraine management to shift from a passive response to proactive prevention and personalized intervention, thereby significantly improving patients' quality of life.

[0092] In one possible design, such as Figure 2 As shown, in order to examine the fluctuation range of underlying lightweight physiological characteristics in the preliminary fusion data within a preset time window, this application may further include the following steps:

[0093] S101. Continuously track individual physiological baselines.

[0094] Among them, continuously tracking an individual's physiological baseline refers to the system's long-term collection and analysis of an individual's physiological data, such as heart rate, respiratory rate, and skin conductance.

[0095] S102. Calculate the instantaneous values ​​of the underlying lightweight physiological characteristics in the preliminary fused data.

[0096] As one possible implementation, the system can extract key physiological indicator values ​​at the current moment from the preliminary fusion data received in real time, as instantaneous values ​​of the underlying lightweight physiological characteristics.

[0097] It should be noted that these lightweight features are usually indicators that are easy to obtain in real time and have some indicative significance for migraine aura or onset, such as heart rate variability, respiratory rhythm, and micro-fluctuations in body temperature.

[0098] S103. Compare the instantaneous value with the individual's physiological baseline to obtain the relative deviation value.

[0099] As one possible implementation, the system can calculate percentage deviation, standard deviation multiple, or Z-score, and use the percentage deviation, standard deviation multiple, or Z-score as relative deviation values.

[0100] S104. Adjust the preset trend threshold and preset fluctuation threshold based on the relative deviation value and the instantaneous value of the prediction confidence.

[0101] As one possible implementation, the system may appropriately relax the threshold when an individual's physiological deviation is large or the prediction confidence is low, in order to more sensitively capture potential anomalies; conversely, when the physiological state is stable and the prediction confidence is high, the threshold may be tightened to reduce false alarms.

[0102] In some preferred embodiments, a specific example is given below. Assume a user wears a physiological sensor, and the system continuously collects data such as heart rate and heart rate variability (HRV), establishing a baseline heart rate of 65-75 beats / minute and an HRV baseline of 50-80 ms. At a certain point in time, the system receives preliminary fused data. After evaluation by the first-level analysis module, a prediction confidence score of 0.65 is obtained, which falls between a first threshold (e.g., 0.7) and a second threshold (e.g., 0.5). At this point, the system needs further monitoring. The system first calculates the instantaneous heart rate at the current moment as 85 beats / minute and the instantaneous HRV as 40 ms. Compared to the individual's physiological baseline, the relative deviation of the heart rate is high (deviating from the normal range), and the relative deviation of the HRV is low (below the normal range). Simultaneously, the prediction confidence score of 0.65 is at a moderate level. Based on this information, the system dynamically adjusts the preset trend threshold and preset fluctuation threshold. For example, if the original preset fluctuation threshold was triggered only when HRV decreased by 20%, now, due to the large relative deviations of heart rate and HRV and low prediction confidence, the system might adjust the preset HRV fluctuation threshold to trigger only when it decreases by 15%, in order to improve sensitivity and avoid missing potential physiological abnormalities. Subsequently, the system uses these adjusted thresholds to determine the trend of prediction confidence changes and the fluctuation amplitude of underlying lightweight physiological characteristics, thereby more accurately deciding whether to initiate the pre-deep analysis process.

[0103] Through the above technical solution, the system overcomes the shortcomings of traditional fixed threshold judgments, significantly improving the accuracy of early identification of migraine symptom precursors or attacks. Because the threshold is dynamically adjusted based on individual physiological baselines and real-time prediction confidence, it better adapts to individual physiological diversity and state changes, reducing false alarms or missed alarms caused by individual differences. This not only makes the allocation of computing resources more precise and efficient, avoiding unnecessary in-depth analysis, but also improves the intelligence and personalization of the entire data processing method, providing users with a more reliable and customized migraine risk assessment.

[0104] In one possible design, in order to initiate the pre-depth analysis process, this application further includes the following steps:

[0105] S201. Perform multi-dimensional feature extraction on the preliminary fused data to obtain multi-dimensional features.

[0106] As one possible implementation, the system can perform in-depth analysis of data from multiple perspectives, including the time domain, frequency domain, time-frequency domain, and nonlinear dynamics, to obtain multi-dimensional features.

[0107] For example, it can extract time-domain statistical features such as mean, variance, kurtosis, and skewness of signals; frequency-domain features such as power spectral density, dominant frequency, and bandwidth; and nonlinear dynamic features such as Lyapunov exponent and fractal dimension.

[0108] S202. Check whether the multidimensional features are accompanied by synchronous changes in other physiological modalities, and obtain the synchronous change situation.

[0109] Crop point clusters refer to dense collections of points formed in a point cloud by one or more crop stems and leaves.

[0110] As one possible implementation, the system cross-compares data from different physiological sensors (such as heart rate sensors, skin conductance sensors, electromyography sensors, body temperature sensors, etc.) to determine whether there is a cross-modal synchronous response.

[0111] For example, if a specific pattern of change in heart rate variability is detected, the system will simultaneously check whether other physiological indicators such as skin conductance, respiratory rate, or muscle activity also show synchronous changes associated with it.

[0112] S203. Adjust the judgment priority of forgery recognition according to the synchronous changes.

[0113] As one possible implementation, when a significant synchronous change in multidimensional features is detected across different physiological modalities, the system reduces the priority of identifying that pattern as a simple artifact and instead tends to treat it as a genuine physiological event. Conversely, if an anomalous pattern appears only in a single modality and lacks synchronous support from other modalities, its priority for being identified as an artifact will be increased accordingly.

[0114] For example, if heart rate and breathing patterns show specific stress-related changes simultaneously, the system will interpret this as more likely a genuine physiological stress response than noise from a single sensor.

[0115] S204. Maintain a dynamically updated physiological pattern feature library.

[0116] As one possible implementation, when a new pattern is identified, the system can perform a preliminary physiological correlation assessment based on the duration, intensity change rate, and synchronicity with other physiological modalities of the new pattern to obtain a preliminary assessment result; and based on the preliminary assessment result, check whether the new pattern is similar to the characteristic patterns of known non-migraine physiological events to obtain a similarity check result.

[0117] Furthermore, when the similarity check results show that the new pattern is highly similar to the characteristic patterns of non-migraine physiological events, the new pattern is marked as a non-migraine-related pattern and stored in the non-migraine pattern temporary storage area; when the similarity check results show that the new pattern is not similar to the characteristic patterns of non-migraine physiological events, a multi-round expert verification process is initiated to independently evaluate the new pattern, obtain expert verification results, and, based on the expert verification results, include the new pattern in the dynamically updated physiological pattern feature library.

[0118] Known patterns of non-migraine physiological events may include, but are not limited to, changes in sleep cycles, motion artifacts, environmental noise interference, drug reactions, and physiological fluctuations caused by other diseases (such as colds and indigestion). Similarity checks can be performed using pattern matching algorithms (such as Dynamic Time Warping (DTW), Support Vector Machines (SVM), and neural networks).

[0119] S205. Compare the identified patterns with the physiological pattern feature library to obtain the comparison results.

[0120] As one possible implementation, the system can use pattern matching algorithms (such as distance-based matching, classifier matching, or deep learning models) to compare the currently extracted multi-dimensional features with known patterns in the feature library and quantify their similarity or attribution probability.

[0121] S206. Determine the pattern type based on the comparison results.

[0122] As one possible implementation, the system can determine that the identified pattern is the preset pattern if the quantified similarity or attribution probability between the identified pattern and the preset pattern is greater than the corresponding set threshold.

[0123] The preset modes can be early signs of migraine, stress response, sleep state, or specific types of artifacts.

[0124] Through the aforementioned technical solutions, the accuracy and robustness of the pre-deep analysis process have been significantly improved. Specifically, by extracting multi-dimensional features and validating cross-modal synchronous changes, the system can more reliably identify early, subtle physiological indicators of migraines and effectively distinguish them from various non-physiological artifacts or benign physiological fluctuations. This enhanced discriminative ability significantly reduces the false alarm rate, avoiding unnecessary resource allocation to the second-level analysis module due to non-migraine-related events. Simultaneously, it minimizes the false negative rate, ensuring that even weak physiological abnormalities are not misclassified as artifacts, thus guaranteeing timely and accurate in-depth analysis of potential migraine symptoms. Therefore, the overall efficiency and diagnostic accuracy of the migraine symptom data processing method have been significantly improved.

[0125] In some preferred embodiments, the system initiates a pre-deep analysis process when it detects a moderate level of prediction confidence (e.g., below a first threshold but above a second threshold) and minimal fluctuations in underlying lightweight physiological characteristics (such as heart rate and respiratory rate). Specifically, the system first extracts multi-dimensional features from the preliminary fused data. For example, it extracts frequency domain features (such as high-frequency power and low-frequency power) of heart rate variability from electrocardiogram (ECG) signals, the amplitude and frequency of fluctuations from skin conductance signals, and time-series features of muscle tension from electromyography (EMG) signals. Subsequently, the system checks whether these multi-dimensional features are accompanied by synchronous changes in other physiological modalities. For example, if a specific pattern of heart rate variability changes occurs synchronously with a sustained increase in skin conductance and a slight increase in neck EMG signals, the system considers this a possible sign of genuine physiological stress or early migraine. In this case, the priority of artifact detection is adjusted to reduce the likelihood of it being identified as an artifact. The system further compares these identified patterns with a dynamically updated physiological pattern feature library. This feature library may contain known motion artifact patterns, sensor noise patterns, and early migraine physiological patterns and stress response patterns verified by experts. If the comparison results show that the current pattern is highly similar to an early migraine physiological pattern and has a low similarity to known artifact patterns, the system will determine that the pattern is a genuine sign of physiological abnormality and further confirm that additional computing resources need to be allocated to call the second-level analysis module for more in-depth analysis.

[0126] In one possible design, in order to initiate a multi-round expert verification process, independently evaluate the new model, and obtain expert verification results, this application further includes the following steps:

[0127] S301. Distribute the new model to multiple experts for independent evaluation and obtain the evaluation results from each expert.

[0128] As one possible implementation, the system can distribute the new model to the terminal devices of multiple experts. Experts can input their evaluation results on their terminal devices and upload them to the system. In turn, the system obtains the evaluation results of each expert.

[0129] S302. When there are inconsistencies in the evaluation results of various experts, the expert verification results shall be obtained based on the expert discussion meeting.

[0130] As one possible implementation, the system can determine that there is inconsistency in the evaluation results of various experts when the maximum proportion of the same result in the evaluation results is less than the proportion threshold, and obtain the expert verification results based on the expert discussion meeting.

[0131] The percentage threshold can be set as needed. For example, it can be set to 80%.

[0132] For example, if the evaluation results show that 75% of the new models are model a, 5% are model b, and 20% are model c, then the maximum percentage of the same result is 75%, which is less than 80%, indicating that there is inconsistency in the evaluation results of the experts.

[0133] In one possible design, in order to obtain expert verification results based on expert discussion meetings, this application also includes:

[0134] S401. Initiate an expert discussion meeting.

[0135] As one possible implementation, the system can send a meeting notification to each expert's terminal device to initiate an expert discussion meeting.

[0136] Meeting notices may include the meeting time, meeting location, etc.

[0137] S402. In the expert discussion meeting, present the multi-dimensional data analysis results of the new model, the expert evaluation records of historical similar models, and relevant clinical guidelines.

[0138] The multi-dimensional data analysis results include signal spectral characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics; the expert evaluation records include the evaluation time, evaluation experts, evaluation conclusions, and evaluation basis.

[0139] The relevant clinical guidelines include diagnostic criteria for migraine, key points for differential diagnosis, and recommendations for treatment interventions.

[0140] As one possible implementation, the system can build an interactive data visualization interface, which can automatically highlight or provide specific dimensions of interpretive auxiliary information based on the expert's preset focus or professional field.

[0141] The interactive data visualization interface displays the multi-dimensional data analysis results of the new mode in real time. The multi-dimensional data analysis results include signal spectrum characteristics, time domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics. The interactive data visualization interface allows experts to dynamically scale, filter, and cross-compare the signal spectrum characteristics, time domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics.

[0142] Among them, signal spectral characteristics can be understood as the analysis of the energy distribution of physiological signals at different frequencies, such as through Fourier transform and other methods; time-domain statistical characteristics refer to the description of the statistical features of physiological signals in time series, such as mean, variance, peak value, waveform duration, etc.; nonlinear dynamic characteristics focus on the inherent complexity and nonlinear behavior of physiological systems, such as through methods like Poincaré diagrams and fractal dimension; and cross-modal correlation characteristics aim to reveal whether there are synchronous or coordinated changes between different physiological modalities (such as heart rate, respiration, EEG, etc.).

[0143] For example, experts can dynamically scale signal spectral characteristics to observe subtle changes within a specific frequency range; they can filter time-domain statistical characteristics to focus on data fluctuations over a specific time period; and they can cross-compare nonlinear dynamic characteristics and cross-modal correlation characteristics to discover potential correlations between different dimensions. These operations aim to improve the efficiency and depth of experts' understanding of complex data.

[0144] S403, based on multi-dimensional data analysis results of the new model, expert evaluation records of historical similar models, and relevant clinical guidelines, assists experts in reaching a consensus on evaluation results and obtains expert-verified results.

[0145] This application's approach provides structured, multi-dimensional data support during expert discussion meetings, enabling experts to engage in discussions on a shared and well-informed basis. Multi-dimensional data analysis results offer experts a comprehensive and in-depth technical perspective, helping them understand new patterns from the essential characteristics of the signals. Historical expert evaluation records of similar patterns provide valuable experiential references, allowing experts to draw upon the judgment logic and basis of predecessors. Simultaneously, the introduction of relevant clinical guidelines ensures that expert discussions and judgments strictly adhere to the latest medical standards and practices, thereby avoiding subjective assumptions or experience biases. Thus, with a full grasp of the information, experts can communicate, debate, and weigh options more effectively, ultimately reaching an evaluation result based on objective evidence and professional consensus.

[0146] In some preferred embodiments, suppose the system identifies a new physiological pattern, but multiple experts disagree on whether it is a migraine-related pattern after independent evaluation. In this case, the system initiates an expert discussion meeting. The meeting first presents detailed, multi-dimensional data analysis results of the new pattern. For example, spectral analysis of its EEG signals shows abnormal enhancement in specific frequency bands, heart rate variability analysis reveals significant fluctuations in autonomic nervous activity, and environmental detector data shows a sudden drop in air pressure prior to the pattern's appearance. Next, the system presents cases with similar signal characteristics to the new pattern from a historical database, along with the experts' assessments and detailed justifications for these cases at the time. Furthermore, the meeting also presents the latest international migraine diagnostic criteria (such as ICHD-3) regarding the differential diagnosis of migraine with aura and migraine without aura, as well as the clinical characteristics of non-migraine physiological events (such as transient ischemic attacks) that may be associated with the pattern. Based on this comprehensive and structured information, experts can engage in in-depth discussions on the new pattern's signal spectral characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics, and conduct comparative analysis in conjunction with historical experience and clinical guidelines. For example, one expert might point out that while some features of the pattern are similar to migraines, its cross-modal associations are more consistent with a known non-migraine physiological event. Another expert might, based on clinical guidelines, emphasize that the duration of the pattern or accompanying symptoms does not fully meet the typical diagnostic criteria for migraine. Through such evidence-based discussions, experts can ultimately reach a consensus, for example, classifying the new pattern as a rare variant of migraine aura or as a non-migraine-related physiological event requiring further observation, thus obtaining a unified and convincing expert-verified result.

[0147] In some of the embodiments described above in this application, in order to automatically highlight or provide specific dimensions of interpretive auxiliary information based on the expert's preset focus or area of ​​expertise, the following are included:

[0148] S501, an interactive data visualization interface that identifies the focus or professional field preset by experts.

[0149] As one possible approach, the system can determine the physiological modality, signal type, analysis dimension, or medical specialty that the expert is currently focusing on by analyzing the expert's operating habits, historical query records, personal configurations, or explicit inputs on the interactive data visualization interface.

[0150] For example, an expert might presuppose a focus on heart rate variability analysis, or their area of ​​expertise might be labeled as neurology.

[0151] S502. An interactive data visualization interface assesses the degree of match between the focus or area of ​​expertise and the actual physiological performance of the new model.

[0152] The system can compare the typical physiological feature patterns corresponding to the points of interest or professional fields identified by experts with the actual multi-dimensional data analysis results of the new pattern to be analyzed, and obtain the degree of matching. The evaluation of this degree of matching can be based on various algorithms, such as feature vector distance calculation, pattern similarity analysis, or machine learning model prediction, to quantify the correlation or consistency between the two.

[0153] First, the interactive data visualization interface will identify the expert's preset focus or area of ​​expertise.

[0154] Experts can explicitly specify their area of ​​expertise (such as neurology or cardiology) in the system settings.

[0155] The system can learn from the expert's historical operating habits on the interface and focus on the physiological feature patterns that are viewed most frequently.

[0156] The signal spectrum characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics of the physiological feature pattern corresponding to the focus are compared with the signal spectrum characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics of the new pattern by cosine similarity calculation. Multiple feature similarities are obtained, and the mean of multiple feature similarities is determined as the matching degree between the focus and the actual physiological performance of the new pattern in the interactive data visualization interface.

[0157] Alternatively, the signal spectrum characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics corresponding to the preset physiological characteristic mode of the professional field can be compared with the signal spectrum characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics of the new mode to calculate the cosine similarity, resulting in multiple characteristic similarities. The mean of the multiple characteristic similarities is then determined as the matching degree between the actual physiological performance of the professional field and the new mode.

[0158] The formula for calculating cosine similarity can be found in existing technologies and will not be elaborated here.

[0159] S503. When the matching degree is lower than the preset matching threshold, the interactive data visualization interface prompts the expert that there is a potential bias.

[0160] As one possible approach, when the matching degree is lower than a preset matching threshold, the system can proactively inform experts that there may be blind spots or misunderstandings in the current analysis through visual warnings, text prompts, or voice reminders, prompting experts to re-examine their focus and alerting them to potential biases.

[0161] The preset matching threshold can be set as needed and is not restricted.

[0162] S504. The interactive data visualization interface recommends supplementary information for interpretation of other relevant dimensions or provides cross-domain interpretation guidance based on potential biases, in order to guide experts in understanding new patterns.

[0163] As one possible implementation, the system can determine the cosine similarity between the signal spectrum characteristics of each physiological feature pattern and the signal spectrum characteristics of the new pattern from the physiological feature database, and recommend the interpretation auxiliary information corresponding to the physiological feature pattern with the highest cosine similarity.

[0164] It should be noted that the physiological pattern feature database pre-stores different physiological feature patterns and corresponding interpretation auxiliary information.

[0165] For example, if experts are focused on cardiovascular health, but the new pattern shows significant EEG abnormalities, the system may recommend neurophysiological interpretation guidelines or provide interdisciplinary expert advice to help experts understand the potential triggers or clinical significance of the new pattern from a more comprehensive perspective.

[0166] Through the aforementioned technical solution, this application can significantly improve the efficiency and accuracy of experts in analyzing complex physiological data. This solution, through intelligent assessment and dynamic guidance, effectively compensates for the information blind spots and comprehension biases that may arise from relying solely on static presets, ensuring that experts receive auxiliary information that highly aligns with the actual performance of the new pattern. Therefore, it not only helps experts more quickly identify the true physiological abnormalities of the new pattern but also promotes the integration of interdisciplinary knowledge, providing stronger technical support for the accurate diagnosis and etiological insight of migraine symptoms, ultimately improving the reliability of overall diagnosis and the scientific rigor of clinical decision-making.

[0167] In some preferred embodiments, a specific example is given below. Suppose a neurologist is using an interactive data visualization interface to analyze a newly identified physiological pattern. The neurologist's primary focus is on the spectral characteristics of electroencephalograms (EEGs) to look for specific rhythmic changes associated with migraines.

[0168] First, the interactive data visualization interface identifies the expert's pre-defined focus as EEG spectral characteristics. Then, the system begins to assess the degree of match between this focus and the actual physiological performance of the new model. Assume that the multi-dimensional data analysis of the new model shows, in addition to slight abnormalities in the EEG spectrum, significant abnormalities in heart rate variability (HRV) and dramatic fluctuations in skin conductance (GSR), characteristics not fully covered in the expert's pre-defined EEG spectral focus.

[0169] At this point, the system-calculated matching degree is lower than the preset matching threshold. The interactive data visualization interface will immediately display a warning message on the screen, such as "Note: The current pattern may involve non-EEG-dominated physiological abnormalities. Please consider other modal data," indicating that there is a potential bias in the expert's analysis.

[0170] Based on this potential bias, the interactive data visualization interface no longer simply highlights EEG spectral data. Instead, it intelligently recommends viewing the temporal statistical and nonlinear dynamic characteristics of HRV and GSR, and provides cross-disciplinary guidance on the association between autonomic nervous system dysfunction and migraine. For example, the system might recommend a clinical literature abstract on "the role of autonomic dysfunction in migraine attacks" or highlight areas of abnormal low-frequency / high-frequency ratios of HRV in the new pattern. In this way, experts are guided to focus on more comprehensive physiological indicators, thereby more accurately understanding the migraine triggers that the new pattern may indicate, such as whether it is related to autonomic nervous system disorders caused by stress or emotional fluctuations, rather than simply focusing on EEG activity.

[0171] like Figure 3 As shown in the figure, this embodiment of the invention also provides a migraine symptom data processing system. The system includes:

[0172] The data receiving and fusion module is used to receive data streams from physiological sensors and environmental detectors, and to perform preliminary fusion processing on the data streams to obtain preliminary fused data;

[0173] The first-level analysis module is used to input the preliminary fused data into the first-level analysis module. The first-level analysis module evaluates the preliminary fused data and obtains the prediction results and prediction confidence.

[0174] The resource allocation module is used to maintain the current computing resources and output the prediction results when the prediction confidence is greater than the first threshold.

[0175] When the prediction confidence is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to call the second-level analysis module;

[0176] The second-level analysis module is used to extract features and recognize patterns from the preliminarily fused data using the allocated computing resources, so as to obtain prediction results and causal insights. Causal insights include explanatory information for the prediction results.

[0177] This application also provides a computer-readable storage medium. All or part of the processes in the above method embodiments can be executed by a computer program instructing related hardware. This program can be stored in the computer-readable storage medium, and when executed, it can include the processes of the above method embodiments. The computer-readable storage medium can be an internal storage unit of the task execution device (including a data sending end and / or a data receiving end) of any of the foregoing embodiments, such as the hard disk or memory of the task execution device. The computer-readable storage medium can also be an external storage device of the terminal device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the terminal device. Further, the computer-readable storage medium can include both the internal storage unit of the task execution device and an external storage device. The computer-readable storage medium is used to store the computer program and other programs and data required by the task execution device. The computer-readable storage medium can also be used to temporarily store data that has been output or will be output.

[0178] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0179] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, essentially, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This software product is stored in a storage medium and includes several instructions to cause a device (which may be a microcontroller, chip, etc.) or processor to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0180] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A method for processing migraine symptom data, characterized in that, include: The system receives data streams from physiological sensors and environmental detectors, and performs preliminary fusion processing on the data streams to obtain preliminary fused data. The preliminary fusion data is input into the first-level analysis module, which evaluates the preliminary fusion data to obtain prediction results and prediction confidence. When the prediction confidence level is greater than the first threshold, maintain the current computing resources and output the prediction result; When the prediction confidence is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to call the second-level analysis module; The second-level analysis module uses the allocated computing resources to perform feature extraction and pattern recognition on the preliminary fused data to obtain prediction results and causal insights, wherein the causal insights include explanatory information for the prediction results. The first-level analysis module uses machine learning models or heuristic rules with low computational complexity; the second-level analysis module uses deep learning techniques or complex machine learning models. When the predicted confidence level is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to invoke the second-level analysis module, including: Monitor the trend of the prediction confidence within a preset time window, and examine the fluctuation range of the underlying lightweight physiological characteristics in the preliminary fusion data within the preset time window; Determine whether the trend of the predicted confidence level change is lower than a preset trend threshold, and whether the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than a preset fluctuation threshold; When the trend of the predicted confidence level is lower than the preset trend threshold and the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than the preset fluctuation threshold, the pre-deep analysis process is initiated; the pre-deep analysis process identifies non-physiological artifact patterns in the preliminary fusion data; When the pre-deep analysis process confirms the presence of real physiological abnormalities, additional computing resources are allocated to invoke the second-level analysis module.

2. The method for processing migraine symptom data according to claim 1, characterized in that, The monitoring of the change trend of the prediction confidence within a preset time window, and the examination of the fluctuation range of the underlying lightweight physiological characteristics in the preliminary fusion data within the preset time window, include: Continuously track individual physiological baselines; Calculate the instantaneous values ​​of the underlying lightweight physiological features in the preliminary fused data; The instantaneous value is compared with the individual's physiological baseline to obtain the relative deviation value; Based on the relative deviation value and the instantaneous value of the prediction confidence, adjust the preset trend threshold and the preset fluctuation threshold; The adjusted preset trend threshold and preset fluctuation threshold are used to determine the trend of the predicted confidence level change and the fluctuation amplitude of the underlying lightweight physiological characteristics.

3. The method for processing migraine symptom data according to claim 1, characterized in that, When the predicted confidence change trend is lower than the preset trend threshold and the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than the preset fluctuation threshold, the pre-deep analysis process is initiated, including: Multi-dimensional feature extraction is performed on the preliminary fused data to obtain multi-dimensional features; Check whether the multidimensional features are accompanied by synchronous changes in other physiological modalities to obtain the synchronous change situation; Based on the aforementioned synchronous changes, adjust the judgment priority of artifact identification; Maintain a dynamically updated database of physiological pattern features; The identified patterns are compared with the physiological pattern feature library to obtain the comparison results; Based on the comparison results, the pattern type is determined.

4. The method for processing migraine symptom data according to claim 3, characterized in that, The maintenance of the dynamically updated physiological pattern feature library includes: When a new pattern is identified, a preliminary physiological relevance assessment is performed based on the duration, rate of change of intensity, and synchronicity with other physiological modalities of the new pattern, and a preliminary assessment result is obtained. Based on the preliminary assessment results, the similarity check results are obtained by examining whether the new pattern is similar to the characteristic patterns of known non-migraine physiological events. When the similarity check results show that the new pattern is highly similar to the characteristic pattern of the non-migraine physiological event or the similarity is greater than the similarity threshold, the new pattern is marked as a non-migraine-related pattern and stored in the non-migraine pattern temporary storage area. When the similarity check results show that the similarity between the new pattern and the characteristic pattern of the non-migraine physiological event is less than the similarity threshold, a multi-round expert verification process is initiated to independently evaluate the new pattern and obtain the expert verification results. Based on the expert verification results, the new pattern will be incorporated into the dynamically updated physiological pattern feature library.

5. The method for processing migraine symptom data according to claim 4, characterized in that, The process of initiating multiple rounds of expert verification involves independently evaluating the new model to obtain expert verification results, including: The new model was distributed to multiple experts for independent evaluation, and the evaluation results of each expert were obtained. When there are inconsistencies in the evaluation results of the various experts, the expert verification results are obtained based on the expert discussion meeting.

6. The method for processing migraine symptom data according to claim 5, characterized in that, The expert verification results obtained based on the expert discussion meeting include: Initiate an expert discussion meeting; The expert discussion meeting will present the multi-dimensional data analysis results of the new model, the expert evaluation records of historical similar models, and relevant clinical guidelines. The multi-dimensional data analysis results include signal spectral characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics; the expert evaluation record includes evaluation time, evaluation experts, evaluation conclusions, and evaluation basis. The relevant clinical guidelines include diagnostic criteria for migraine, key points for differential diagnosis, and recommendations for treatment interventions. Based on the multi-dimensional data analysis results of the new model, expert evaluation records of historical similar models, and relevant clinical guidelines, the experts reached a consensus on the evaluation results, and the expert verification results were obtained.

7. The method for processing migraine symptom data according to claim 6, characterized in that, The multi-dimensional data analysis results showcasing the new model include: An interactive data visualization interface is constructed, which displays the multi-dimensional data analysis results of the new mode in real time. The multi-dimensional data analysis results include signal spectrum characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics. The interactive data visualization interface allows experts to dynamically zoom, filter, and cross-compare the signal spectral characteristics, time-domain statistical characteristics, nonlinear dynamic characteristics, and cross-modal correlation characteristics. The interactive data visualization interface automatically highlights or provides specific dimensions of interpretive assistance information based on the expert's preset focus or area of ​​expertise.

8. A method for processing migraine symptom data according to claim 7, characterized in that, The interactive data visualization interface automatically highlights or provides specific dimensions of interpretive aids based on the expert's preset focus or area of ​​expertise, including: The interactive data visualization interface identifies the expert's preset focus or area of ​​expertise. The interactive data visualization interface assesses the degree of match between the focus or area of ​​expertise and the actual physiological performance of the new pattern. When the matching degree is lower than a preset matching threshold, the interactive data visualization interface prompts the expert that there is a potential bias; The interactive data visualization interface recommends supplementary information for interpretation in other relevant dimensions or provides cross-domain interpretation guidance based on the potential deviations, in order to guide experts in understanding the new pattern.

9. A migraine symptom data processing system, characterized in that, The system includes: The data receiving and fusion module is used to receive data streams from physiological sensors and environmental detectors, and to perform preliminary fusion processing on the data streams to obtain preliminary fused data; The first-level analysis module is used to input the preliminary fusion data into the first-level analysis module, and the first-level analysis module evaluates the preliminary fusion data to obtain the prediction result and prediction confidence. The resource allocation module is used to maintain the current computing resources and output the prediction result when the prediction confidence is greater than a first threshold. When the prediction confidence is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to call the second-level analysis module; The second-level analysis module is used to extract features and recognize patterns from the preliminary fused data using the allocated computing resources to obtain prediction results and causal insights, wherein the causal insights include explanatory information for the prediction results. The first-level analysis module uses machine learning models or heuristic rules with low computational complexity; the second-level analysis module uses deep learning techniques or complex machine learning models. When the predicted confidence level is less than or equal to the first threshold and greater than or equal to the second threshold, additional computing resources are allocated to invoke the second-level analysis module, including: Monitor the trend of the prediction confidence within a preset time window, and examine the fluctuation range of the underlying lightweight physiological characteristics in the preliminary fusion data within the preset time window; Determine whether the trend of the predicted confidence level change is lower than a preset trend threshold, and whether the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than a preset fluctuation threshold; When the trend of the predicted confidence level is lower than the preset trend threshold and the fluctuation amplitude of the underlying lightweight physiological characteristics is lower than the preset fluctuation threshold, the pre-deep analysis process is initiated; the pre-deep analysis process identifies non-physiological artifact patterns in the preliminary fusion data; When the pre-deep analysis process confirms the presence of real physiological abnormalities, additional computing resources are allocated to invoke the second-level analysis module.

Images