Multi-modal fusion csvd cognitive function dynamic monitoring and early warning system

Abstract

The present application relates to the technical field of medical monitoring and intelligent early warning, in particular to a multi-modal fusion CSVD cognitive function dynamic monitoring and early warning system, which solves the problem of dynamic, accurate and comprehensive monitoring and early warning of cognitive function and emotional disorders of CSVD patients in clinical practice. It comprises a multi-modal data acquisition module for collecting and acquiring multi-dimensional data of patients; a data preprocessing module for standardizing data; a multi-modal fusion analysis module for extracting cognitive and emotional features, constructing a model and generating a fusion feature vector; a cognitive function dynamic monitoring module for real-time evaluation of cognitive function; an emotional state evaluation module for identifying emotional state risks and outputting results; a risk early warning module for grading early warning and suggestions; a storage interaction module for supporting data storage and docking; and a terminal display module for visualizing monitoring, evaluation and early warning results. Through the cooperation of multiple modules, the present application realizes comprehensive monitoring and accurate early warning of the cognitive and emotional state of CSVD patients, and improves the pertinence and accuracy of clinical application.

Description

Technical Field

[0001] This invention relates to the field of medical monitoring and intelligent early warning technology, specifically to a multimodal fusion-based dynamic monitoring and early warning system for cognitive function in patients with CSVD (Cognitive Disorders). Background Technology

[0002] Cerebral small vessel disease (CSVD) is a group of clinical imaging syndromes affecting intracranial arterioles, capillaries, and other microvessels. Typical imaging manifestations include white matter hyperintensity, lacunar infarction, and microbleeds. It has a high incidence, insidious course, and slow progression, and is the most common cause of vascular cognitive impairment, accounting for 36% to 67% of vascular dementia cases. As CSVD progresses, it can lead to cognitive impairment, affective disorders, and other neurological deficits. Affective disorders often manifest as depression and apathy, and are an independent risk factor for depression in the elderly. Cognitive impairment is the main cause of vascular dementia, severely reducing patients' quality of life. Early detection and intervention are crucial for reducing the social burden and improving patient prognosis.

[0003] Currently, the monitoring of cognitive and emotional disorders in CSVD mainly relies on single-modal methods, such as neuroimaging examinations (head MRI, CT), clinical cognitive / emotional assessment scales, and laboratory biomarker detection. However, existing technologies have obvious limitations: First, single-modal monitoring is one-sided. For example, imaging examinations have low sensitivity and are difficult to capture subtle changes in early cognitive function and emotional state. Laboratory tests have insufficient application of specific biomarkers for CSVD-related emotional and cognitive disorders. Clinical scale assessments are highly subjective and easily influenced by the experience of the testers, resulting in a high rate of missed diagnosis and misdiagnosis. Secondly, existing monitoring methods are mostly static, phased detection methods, which cannot achieve dynamic tracking of cognitive function and emotional state, and are difficult to reflect the trend of disease progression in real time. Thirdly, there is a lack of in-depth fusion and analysis capabilities for multi-source data, especially the failure to fully integrate serum inflammatory immune markers (such as the neutrophil-to-lymphocyte ratio NLR, interleukin IL-17A, nitric oxide NO, etc.) with other modal data. This makes it impossible to coordinate the interpretation of multi-dimensional data such as imaging features, physiological signals, inflammatory immune markers, and clinical information, resulting in insufficient accuracy and delayed warnings, and failing to provide timely support for early intervention.

[0004] Existing research has confirmed that the occurrence and development of chronic cerebral hypoperfusion (CSVD) are closely related to chronic brain hypoperfusion, blood-brain barrier damage, endothelial dysfunction, and neuroinflammatory responses. Among these, the inflammatory immune response plays a crucial role in the pathological process of CSVD. Serum inflammatory immune marker levels are closely related to emotional and cognitive impairments in CSVD patients. Elevated IL-17A and NLR are independent risk factors for emotional and cognitive impairments in CSVD patients, while elevated NO is an independent protective factor. However, current technologies have not effectively integrated these key research findings into monitoring and early warning systems, failing to achieve the synergistic application of inflammatory immune markers with other multimodal data. Currently, a deep integration of multimodal data with CSVD cognitive function monitoring has not been achieved, failing to meet the clinical need for dynamic, accurate, and comprehensive monitoring and early warning of CSVD cognitive function and emotional disorders. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a dynamic monitoring and early warning system for CSVD cognitive function that integrates multi-dimensional monitoring data to achieve real-time dynamic monitoring, accurate assessment, and early risk warning of CSVD patients' cognitive function, while also incorporating multimodal fusion of monitoring equipment integration and intelligent medical data processing.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution: a multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function, comprising a multimodal data acquisition module, a data preprocessing module, a multimodal fusion analysis module, a cognitive function dynamic monitoring module, an emotional state assessment module, a risk early warning module, a storage and interaction module, and a terminal display module, wherein each module is electrically connected in sequence to collaboratively realize the dynamic monitoring and early warning of CSVD cognitive function and emotional state; The multimodal data acquisition module includes imaging data, neurophysiological signal data, serum inflammatory and immune biomarker data, clinical cognitive assessment data, clinical emotional assessment data, and lifestyle behavior data. The module comprises an integrated monitoring terminal and external acquisition devices. The integrated monitoring terminal has built-in ECG sensors, blood oxygen sensors, and sleep monitoring sensors. The external acquisition devices include a cranial MRI device, a blood testing device, a cognitive assessment terminal, and an emotional assessment terminal. Serum inflammatory and immune biomarker data is collected through the blood testing device, specifically including the concentration data of neutrophil-to-lymphocyte ratio (NLR), nitric oxide (NO), superoxide dismutase (SOD), interleukin-17A (IL-17A), and interferon-α (IFN-α). Compared with existing technologies, this module adds core detection indicators for serum inflammatory and immune biomarkers, aligning with clinical research conclusions. The data preprocessing module is used to clean, standardize, denoise, and align the raw data collected by the multimodal data acquisition module, remove abnormal data, fill in missing values, and convert data of different formats and dimensions into structured data of a unified standard to generate a preprocessed data set.

[0007] The multimodal fusion analysis module is used to perform multimodal fusion analysis on the preprocessed dataset, extract the cognitive function and emotional state correlation features from each modality of data, and combine the correlation conclusions between serum inflammatory immune markers and CSVD emotion and cognitive impairment to construct a multimodal fusion cognitive-emotion joint assessment model. The multimodal fusion analysis module adopts the CNN-LSTM fusion algorithm. First, the static features in the imaging data and serum inflammatory immune marker data are extracted through the CNN network, and the temporal dynamic features in the neurophysiological signal data and daily behavior data are extracted through the LSTM network. Then, the static features and dynamic features are weighted and fused through the attention mechanism to generate a fused feature vector.

[0008] The cognitive function dynamic monitoring module is used to assess the cognitive function status of CSVD patients in real time and track the trend of cognitive function changes based on the fusion feature vector generated by the multimodal fusion analysis module. The cognitive function dynamic monitoring module has a preset cognitive function assessment index system, including four dimensions: memory, attention, executive function, and language function. Each dimension has a corresponding assessment threshold (the normal threshold for memory is a MoCA scale memory sub-item score ≥3, for attention is a MoCA scale attention sub-item score ≥3, for executive function is a MoCA scale executive function sub-item score ≥2, and for language function is an MMSE scale language sub-item score ≥9). By comparing the fusion feature vector with the assessment index system, real-time cognitive function assessment results are generated (divided into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal). At the same time, based on the time-series fusion feature vector, a dynamic change curve of cognitive function is plotted to achieve dynamic tracking of cognitive function. Combined with the characteristics of serum inflammatory immune markers, when the IL-17A and NLR characteristic values ​​exceed the normal range, or the NO characteristic value is lower than the normal range, the risk points of cognitive function abnormality are automatically marked, and the trend of changes in related dimensions is tracked in a focused manner. The emotional state assessment module is used to assess the emotional state of CSVD patients in real time based on fused feature vectors and clinical emotional assessment data, and to identify the risk of developing emotional disorders (mainly depressive symptoms). It employs the Hamilton Depression Rating Scale (HAMD-24) scoring criteria (total score ≤6 indicates no depression, 7-16 indicates possible depression, 17-23 indicates moderate depression, and ≥24 indicates severe depression), combined with serum inflammatory immune marker characteristics (positive correlation between elevated IL-17A and NLR and emotional disorders, and negative correlation between elevated NO and emotional disorders), to construct an emotional state assessment sub-model, generate emotional state assessment results, and simultaneously plot dynamic change curves of emotional state. These curves are then connected and analyzed with dynamic change curves of cognitive function to capture the synergistic changes between the two, aligning with the high comorbidity of emotional and cognitive disorders in CSVD patients in clinical practice. The risk warning module is used to construct a risk warning model based on the assessment results and fusion feature vectors of the cognitive function dynamic monitoring module and the emotional state assessment module, combined with the risk association characteristics of serum inflammatory immune markers, so as to realize the graded warning of cognitive function impairment and emotional disorders in CSVD. The risk warning module presets three warning thresholds (mild risk 30%~50%, moderate risk 50%~70%, severe risk ≥70%). When the risk probability reaches the corresponding warning threshold, a warning signal is automatically generated. The warning signal includes the warning level, the cause of the risk, and intervention suggestions. The cause of the risk is determined based on the contribution analysis of each modality fusion feature, with a focus on clarifying the abnormal impact of serum inflammatory immune markers (IL-17A, NLR, NO).

[0009] The storage interaction module stores preprocessed datasets, fused feature vectors, cognitive function assessment results, emotional state assessment results, early warning records, and system parameters. It also provides a data interaction interface to support integration with hospital electronic medical record systems and clinical diagnosis and treatment systems, enabling data sharing and synchronous updates. The module employs a distributed storage architecture (storage capacity no less than 10TB), supporting real-time data writing and rapid querying (query response time ≤1s). It also features data encryption (using AES-256 encryption algorithm) to ensure patient privacy and data security. A separate serum inflammatory immune biomarker database is established, linked to patient cognitive and emotional assessment results, forming a longitudinal monitoring dataset for subsequent model optimization and reuse in clinical research. The data interaction interface uses the HL7 protocol for seamless integration with relevant hospital systems, supporting bidirectional synchronous data updates. The terminal display module is used to visually present the dynamic monitoring results of cognitive function, the assessment results of emotional state, the risk warning signals, and the fusion analysis report to medical staff and patients. This includes dynamic curves of cognitive function and emotional state, warning level prompts, details of assessment indicators, and a list of intervention suggestions. The terminal display module supports multi-terminal adaptation, including doctor workstations, patient mobile devices, and nursing terminals. For medical staff terminals, it additionally displays the specific detection values, trends, and correlation analysis reports of serum inflammatory immune markers with cognitive and emotional states, providing accurate references for clinical intervention. The display layout is optimized for different terminals: the PC displays complete and detailed data, the mobile APP displays simplified content, and the tablet displays key monitoring content. The cognitive function dynamic monitoring module, emotional state assessment module, and risk warning module are connected. When abnormal changes in the patient's cognitive function or emotional state are detected, or when the risk probability reaches the warning threshold, the risk warning module immediately triggers a warning signal. At the same time, the cognitive function dynamic monitoring module and emotional state assessment module jointly generate an abnormal change analysis report, which clarifies the specific dimensions and related factors of the decline in cognitive function and emotional state, and focuses on analyzing the abnormal impact of serum inflammatory immune markers, providing accurate reference for clinical intervention.

[0010] To better realize the present invention, the data collected by the multimodal data acquisition module in each dimension are specifically as follows: Imaging data: acquired using cranial MRI equipment, including the volume of high signal in the white matter, the number of lacunar infarcts, the number and location of cerebral microbleeds, and the width of the perivascular space; Neurophysiological signal data: collected through an integrated monitoring terminal, including electrocardiogram signals, blood oxygen saturation, electroencephalogram signals, and sleep cycle data (sleep onset time, deep sleep duration, light sleep duration, and number of awakenings). Serum inflammatory immune marker data: collected through blood testing equipment, including neutrophil-to-lymphocyte ratio (NLR), nitric oxide (NO), superoxide dismutase (SOD), interleukin-17A (IL-17A), and interferon-α (IFN-α) concentration data. NLR was calculated from the synchronously collected neutrophil and lymphocyte test values. The test time point was synchronized with the cognitive and emotional assessment time point to reduce data bias. Clinical cognitive assessment data: collected through cognitive assessment terminals, including scores from the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA). For those with less than 12 years of education, points were added to their total MoCA score to correct for educational bias. Clinical affective assessment data: collected through affective assessment terminals, including Hamilton Depression Rating Scale (HAMD-24) scores, with a total score ≤6 indicating no depression, 7-16 indicating possible depression, 17-23 indicating moderate depression, and ≥24 indicating severe depression. To better realize the present invention, the training process of the multimodal fusion cognitive-emotion joint evaluation model further includes: 1) Collect multimodal sample data of CSVD patients, including 227 confirmed patients (58 without emotional or cognitive impairment, 55 with isolated emotional impairment, 56 with isolated cognitive impairment, and 58 with comorbid emotional and cognitive impairment) and 50 healthy controls. Label the samples with corresponding cognitive function status labels and emotional status labels. The sample data fits the distribution of clinical research samples to ensure that the model is adapted to the actual clinical scenario. 2) Preprocess the sample data using the same processing methods as the data preprocessing module, focusing on calibrating the detection error of serum inflammatory immune marker data to generate a standardized sample dataset; 3) Divide the standardized sample dataset into training and test sets in a 7:3 ratio. Use the training set to train the CNN-LSTM fusion algorithm. Combine the conclusions of clinical multivariate regression analysis to optimize the feature weights of IL-17A, NLR, and NO, and adjust the model parameters (learning rate, number of iterations, feature weights). 4) When the model's cognitive function state classification accuracy reaches over 90% and recall reaches over 88%, and the emotional state classification accuracy reaches over 89% and recall reaches over 87%, the training of the multimodal fusion cognitive-emotion joint evaluation model is completed, and the trained model is deployed to the multimodal fusion analysis module. The model's classification accuracy (Acc) and recall (Rec) are calculated as follows to quantify model performance:

[0011] Among them, TP represents true positives (the number of samples that are actually abnormal but the model predicts are abnormal), TN represents true negatives (the number of samples that are actually normal but the model predicts are normal), FP represents false positives (the number of samples that are actually normal but the model predicts are abnormal), and FN represents false negatives (the number of samples that are actually abnormal but the model predicts are normal). For multi-classification scenarios (4 dimensions of cognitive function and 4 levels of emotional state), weighted accuracy and weighted recall are used for comprehensive evaluation.

[0012] To better realize the present invention, the integrated monitoring terminal adopts a wearable design, weighs ≤50g, supports 24-hour continuous monitoring, has a battery life of ≥72 hours, and has wireless transmission capabilities (Bluetooth, WiFi), which can transmit the collected neurophysiological signal data to the data preprocessing module in real time; the external acquisition device is connected to the integrated monitoring terminal through a standardized interface, supporting real-time data synchronization; the blood testing device supports fully automated testing, which can transmit the serum inflammatory immune marker detection results to the data acquisition module in real time, with a detection error of ≤5%, meeting clinical testing standards.

[0013] To better realize this invention, the storage interaction module further employs the AES encryption algorithm to encrypt patient privacy data, complying with medical data security standards. It also supports data backup, periodically backing up stored data to a cloud server to prevent data loss. The data interaction interface uses the HL7 protocol, enabling seamless integration with hospital electronic medical record systems and clinical diagnosis and treatment systems. This facilitates medical staff in obtaining complete patient diagnosis and treatment information and longitudinal monitoring data of serum inflammatory and immune biomarkers. The serum inflammatory and immune biomarker database supports searching by detection time and biomarker type, allowing medical staff to quickly view trends in patient indicators.

[0014] Compared with the prior art, the present invention has the following advantages and beneficial effects: (1) This invention integrates multi-dimensional data such as imaging, neurophysiological signals, serum inflammatory and immune markers, clinical assessment, and lifestyle behavior. It focuses on integrating core markers closely related to emotional and cognitive impairment in CSVD, such as IL-17A, NLR, and NO, to comprehensively capture the influencing factors of cognitive function and emotional state in CSVD patients. In particular, it strengthens the monitoring and analysis of inflammatory and immune responses. Compared with single-modal monitoring and traditional multimodal systems, it significantly improves the accuracy and comprehensiveness of cognitive function and emotional state assessment, and reduces the rate of missed diagnosis and misdiagnosis. (2) This invention extracts temporal dynamic features and combines them with the dynamic change curves of cognitive function and emotional state to achieve 24-hour continuous monitoring and long-term dynamic tracking of cognitive function and emotional state of CSVD patients. It can capture subtle changes in early cognitive function and emotional state, and at the same time, combined with the dynamic changes of serum inflammatory immune markers, it can predict abnormal risks in advance and reflect the trend of disease progression in real time. (3) This invention combines the conclusions of clinical multivariate regression analysis to construct a multimodal fusion risk warning model. Based on the three-level warning threshold, it realizes the graded warning of cognitive function impairment and emotional disorders, clarifies the warning level, risk cause and targeted intervention suggestions. The warning response time is short, which can provide medical staff with timely and accurate intervention basis. Especially for patients with mild risk, the disease progression can be delayed by adjusting lifestyle and regularly monitoring biomarker levels. (4) This invention supports seamless integration with hospital electronic medical record systems and clinical diagnosis and treatment systems to achieve multi-terminal data sharing and establish a separate serum inflammatory immune marker database to provide data support for clinical research. At the same time, it uses AES encryption algorithm to ensure the security of patient privacy data. The integrated wearable monitoring terminal and automated blood testing equipment are conveniently designed and adaptable to clinical diagnosis and treatment and home monitoring scenarios, improving the clinical practicality and ease of use of the system. Attached Figure Description

[0015] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a schematic block diagram of the structure of the present invention; Figure 2 This is a schematic block diagram of the multimodal data acquisition module in this invention; Figure 3 This is a schematic block diagram of the data preprocessing structure in this invention; Figure 4 This is a schematic block diagram of the multimodal fusion analysis module in this invention; Figure 5 This is a schematic block diagram of the cognitive function dynamic monitoring module in this invention; Figure 6 This is a schematic block diagram of the emotional state assessment module in this invention; Figure 7 This is a schematic block diagram of the risk warning module in this invention; Figure 8 This is a schematic block diagram of the terminal display module in this invention. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0017] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0018] Example 1: This embodiment provides a multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function, the structure of which is as follows: Figure 1 As shown, it includes a multimodal data acquisition module, a data preprocessing module, a multimodal fusion analysis module, a cognitive function dynamic monitoring module, an emotional state assessment module, a risk warning module, a storage and interaction module, and a terminal display module;

[0019] The multimodal data acquisition module collects raw data from various dimensions of CSVD patients; The data preprocessing module cleans, standardizes, reduces noise, aligns, and calibrates the detection error of the collected multi-dimensional raw data to generate a preprocessed dataset. The multimodal fusion analysis module performs multimodal fusion analysis on the preprocessed dataset, extracts cognitive function and emotional state correlation features from each modality, constructs a multimodal fusion cognitive and emotional joint evaluation model, and generates a fusion feature vector; The cognitive function dynamic monitoring module assesses the cognitive function status of CSVD patients in real time based on fused feature vectors, plots the dynamic change curve of cognitive function, and realizes dynamic tracking of cognitive function status. The emotional state assessment module assesses the emotional state of CSVD patients in real time based on fused feature vectors, identifies the risk of emotional disorder, plots dynamic changes in emotional state, and obtains emotional state assessment results. The risk warning module calculates the risk probability of cognitive impairment and emotional disorders based on the dynamic status of cognitive function, the results of emotional state assessment, and the fusion feature vector, and generates graded warning signals and intervention suggestions based on the three-level warning threshold. The storage interaction module stores relevant data, provides a data interaction interface, and enables connection with the hospital's electronic medical record system and clinical diagnosis and treatment system. The terminal display module presents monitoring results, emotional state assessment results, early warning signals, and intervention suggestions in a visual format.

[0020] The specific implementation process includes the following steps: Step 1: System Deployment and Initialization The system hardware architecture was built, and the multimodal data acquisition module, data preprocessing module, multimodal fusion analysis module, cognitive function dynamic monitoring module, emotional state assessment module, risk warning module, storage interaction module, and terminal display module were electrically connected via wired (Ethernet) or wireless (Bluetooth 5.0, WiFi) methods. The system parameters were initialized: basic parameters such as data acquisition frequency, preprocessing standards, assessment threshold, and warning threshold were set. The trained multimodal fusion cognitive and emotional joint assessment model (cognitive classification accuracy ≥90%, emotional classification accuracy ≥89%) was deployed, and the docking and debugging with the hospital's electronic medical record system and clinical diagnosis and treatment system were completed.

[0021] Step 2: Multi-dimensional raw data collection Through the multimodal data acquisition module, multidimensional raw data of CSVD patients are collected simultaneously, including: imaging data (head MRI scan), neurophysiological signal data (collected 24 hours a day by wearable terminal), serum inflammatory and immune marker data (blood test), clinical cognitive / emotional assessment data (scale completion), and daily living behavior data (uploaded by patients on mobile devices). This ensures that the data covers the core influencing factors of cognitive and emotional disorders in CSVD. Among them, serum inflammatory and immune marker data is collected first to ensure that the detection time is synchronized with the cognitive and emotional assessment time.

[0022] Step 3: Data Preprocessing The data preprocessing module receives raw data from each modality and sequentially performs cleaning (removing outliers), standardization (Z-score method), noise reduction (filtering signal interference), alignment (timestamp alignment), and detection error calibration (focusing on calibrating serum inflammatory and immune biomarker data) to generate a standardized structured dataset. This ensures that the data is free of redundancy and dimensionless, meeting the needs of subsequent fusion analysis. The data missing rate after preprocessing is ≤5%, and the serum biomarker calibration error is ≤3%.

[0023] Step 4: Multimodal fusion analysis The multimodal fusion analysis module performs fusion analysis on the preprocessed dataset. Using the CNN-LSTM fusion algorithm, it extracts static features from imaging data and serum inflammatory immune biomarkers, and extracts temporal dynamic features from neurophysiological signals and daily behavior data. Through attention-based weighted fusion, it generates a fusion feature vector. Using the multimodal fusion cognition and emotion joint assessment model, it outputs preliminary cognitive and emotion assessment reference results, providing input for the subsequent monitoring module.

[0024] Step 5: Dynamic monitoring of cognitive function and emotional state The cognitive function dynamic monitoring module, based on fusion feature vectors, assesses patients' cognitive functions (memory, attention, executive function, and language function) in real time, plots dynamic change curves, and marks abnormal risk points. The emotional state assessment module, based on fusion feature vectors, assesses patients' emotional state, identifies the risk of emotional disorders, plots emotional dynamic curves, connects and analyzes the changing patterns of the two modules, and captures comorbidity risks.

[0025] Step 6: Tiered Risk Warning The risk warning module receives the dynamic status of cognitive function, the results of emotional state assessment, and the fusion feature vector. Combined with the abnormal status of serum inflammatory immune markers, it calculates the risk probability through the risk warning model, compares it with the three-level warning thresholds (mild 30%~50%, moderate 50%~70%, severe ≥70%), generates graded warning signals and targeted intervention suggestions, and pushes them to the terminal display module in real time.

[0026] Step 7: Data Storage and Visualization The storage and interaction module stores all data (raw data, preprocessed data, fusion features, evaluation results, early warning records, etc.) and achieves data sharing with the hospital system through the HL7 protocol; the terminal display module displays various monitoring results, early warning signals, and intervention suggestions in a visual form (curves, charts, text), supports simultaneous viewing on multiple terminals, and ensures that medical staff and patients can obtain information in a timely manner.

[0027] Step 8: System Maintenance and Optimization Regular (once a month) system maintenance is performed, including calibrating data acquisition equipment (especially blood testing equipment) and backing up stored data. Based on new monitoring data and clinical feedback, the parameters of the multimodal fusion model and risk warning model are fine-tuned to continuously improve the accuracy of assessment and warning.

[0028] Example 2:

[0029] This embodiment, based on the above embodiments, further defines the structure of the multimodal data acquisition module, such as... Figure 2 As shown, the multimodal data acquisition module includes an integrated monitoring terminal and an external acquisition device;

[0030] The integrated monitoring terminal has built-in electrocardiogram sensor, blood oxygen sensor, and sleep monitoring sensor to collect neurophysiological signal data; The electrocardiogram sensor collects electrocardiogram signal data; The blood oxygen sensor collects blood oxygen saturation data; The sleep monitoring sensor collects electroencephalogram (EEG) signal data and sleep cycle data; The external data acquisition devices include a cranial MRI machine, a blood testing machine, a cognitive assessment terminal, and an emotion assessment terminal; The cranial MRI equipment acquires imaging data; The blood testing equipment collects serum inflammatory immune marker data; The cognitive assessment terminal collects clinical cognitive assessment data; The emotion assessment terminal collects clinical emotion assessment data.

[0031] The specific implementation process includes the following steps: Step 1: Set up the data acquisition module Assemble a multimodal data acquisition module, including an integrated monitoring terminal and external acquisition devices: The integrated monitoring terminal adopts a wristband-style wearable design (weight ≤50g, battery life ≥72 hours), with built-in ECG sensor, blood oxygen sensor, and sleep monitoring sensor, and supports wireless transmission; the external acquisition devices include a 1.5T cranial MRI device, a fully automated blood testing device, a cognitive assessment terminal, and an emotion assessment terminal. All devices are connected to the integrated monitoring terminal through standardized interfaces to complete data synchronization and debugging.

[0032] Step 2: Data Acquisition from Integrated Monitoring Terminal The integrated monitoring terminal is worn on the patient's wrist and a 24-hour continuous monitoring mode is activated: the ECG sensor collects ECG signal data (heart rate, heart rhythm, etc.) every 100ms, the blood oxygen sensor collects blood oxygen saturation data every 30s, and the sleep monitoring sensor collects EEG signals and sleep cycle data (sleep onset time, deep / light sleep duration, number of awakenings) every 1s. The collected data is transmitted to the data preprocessing module in real time to avoid data loss.

[0033] Step 3: Data acquisition using external acquisition devices (1) Head MRI equipment: The patient's head is scanned once a month to collect imaging data such as the volume of high signal in the white matter and the number of lacunar infarcts. After the scan is completed, the data is imported into the system via wired transmission. (2) Blood testing equipment: Venous blood is collected from patients once a week, and serum inflammatory immune marker concentrations (NLR, NO, IL-17A, etc.) are automatically detected with an error of ≤5%. The test results are transmitted to the system in real time. (3) Cognitive assessment terminal: once every 2 weeks, operated by medical staff to guide patients to complete the MMSE and MoCA scales. The terminal automatically records the scoring data. For those with less than 12 years of education, the MoCA total score is adjusted by adding points. (4) Emotional assessment terminal: once every 2 weeks, synchronous cognitive assessment, guide patients to complete the HAMD-24 scale, the terminal automatically records the scoring data and generates preliminary emotional assessment reference.

[0034] Step 4: Data Synchronization and Verification The system automatically timestamps and synchronizes the data collected by the integrated monitoring terminal and the external acquisition device to ensure that multimodal data at the same time point can be correlated. At the same time, it verifies the integrity of the data. If data is missing (such as serum test data not being collected), the system automatically reminds medical staff to collect the missing data to ensure the integrity and correlation of the collected data.

[0035] Example 3: Based on the above embodiments, this embodiment further defines the structure of the multimodal data acquisition module. The imaging data acquired by the cranial MRI device includes the volume of high signal in the white matter, the number of lacunar infarcts, and the number and location of cerebral microbleeds.

[0036] The serum inflammatory immune marker data collected by the blood testing device include the neutrophil-to-lymphocyte ratio, nitric oxide, superoxide dismutase, interleukin-17A, and interferon-α concentration data. The clinical cognitive assessment data collected by the cognitive assessment terminal includes the scores of the MMSE scale and the MoCA scale. For those with less than 12 years of education, additional points are added to their total MoCA score. The clinical emotion assessment data collected by the emotion assessment terminal includes HAMD-24 scale score data.

[0037] Cognitive assessment terminal (collects clinical cognitive assessment data) Data collection method: The terminal has built-in MMSE and MoCA scales, and the data is collected using the "medical staff operation or standardized data entry" mode. The terminal is pre-loaded with complete items for two scales (such as orientation and memory items for MMSE, and attention and executive function items for MoCA), and supports touch / button operation; Based on the patient's on-site responses, medical staff select the corresponding answers on the terminal (or the patient completes and enters simple questions independently), and the terminal automatically generates the original scale score based on the answers; For those with less than 12 years of education, additional points are added to their total MoCA score. The terminal has a built-in automatic correction function that automatically corrects the MoCA score after the patient's years of education are entered. The scoring data is encrypted and stored in real time, and synchronously transmitted to the multimodal data acquisition module to complete the collection of clinical cognitive assessment data.

[0038] Emotional assessment terminal (collects clinical emotional assessment data) Data collection method: The terminal has a built-in HAMD-24 scale, and a medical staff-guided and tiered data entry mode is used. The terminal has 24 assessment indicators of the HAMD-24 scale (such as depressive mood, sleep disorders, etc.), and each indicator corresponds to a 0-4 level scoring standard (in line with the preferred solution of the invention). Medical staff guide patients to complete the assessment through the terminal interface (for elderly patients or those with slightly impaired cognitive function, medical staff can describe the questions on their behalf), and check the corresponding level according to the patient's performance; The terminal automatically summarizes the scores of 24 indicators, generates the total score of the HAMD-24 scale and a preliminary emotional state classification (no depression, possible depression, etc., corresponding to the invention content). The scoring data is synchronized in real time to the multimodal data acquisition module and connected with other modal data to complete the collection of clinical emotion assessment data.

[0039] The specific implementation process includes the following steps: Step 1: Imaging Data Acquisition and Processing A 1.5T cranial MRI scanner was used to perform brain scans on the patient. The scanning parameters were set as follows: T1-weighted imaging (TR=500ms, TE=15ms) and T2-weighted imaging (TR=3000ms, TE=80ms). The scan range covered the entire brain. After acquisition, the volume of high signal in the white matter, the number of lacunar infarcts, the number and location of cerebral microbleeds, and the width of the perivascular space were automatically extracted by the image processing software. The extracted parameters were converted into structured data and transmitted to the data preprocessing module.

[0040] Step 2: Collection and Calculation of Serum Inflammatory Immunological Markers (1) Collect 5 mL of fasting venous blood from the patient, centrifuge (3000 r / min, centrifuge for 10 min), and separate the serum; (2) The concentrations of NO, SOD, IL-17A and IFN-α in serum were detected using a fully automated blood testing device. The detection methods were colorimetric method (NO), xanthine oxidase method (SOD), and enzyme-linked immunosorbent assay (IL-17A and IFN-α), respectively. (3) Simultaneously detect the patient's neutrophil and lymphocyte counts, calculate the NLR value using the formula NLR = neutrophil count / lymphocyte count, and simultaneously calibrate the detection error of neutrophils and lymphocytes to ensure the accuracy of the NLR value; (4) All serum inflammatory immune marker data (NLR, NO, IL-17A, etc.) are organized into structured data, the detection time is marked, and the data is transmitted to the data preprocessing module.

[0041] Step 3: Clinical Cognitive Assessment Data Collection Trained medical staff operate the cognitive assessment terminal to guide patients in completing the MMSE and MoCA scales. The MMSE scale covers seven dimensions, including orientation, memory, and attention, with a total score of 30. The MoCA scale covers eight dimensions, including attention and executive function, with a total score of 30. For patients with less than 12 years of education, 1 point is added to the total MoCA score to correct for educational bias. The terminal automatically records the scores of each sub-item and the total score and transmits them to the system.

[0042] Step 4: Clinical Emotional Assessment Data Collection Synchronous cognitive assessment: Medical staff guide patients to complete the HAMD-24 scale, which covers 24 items including depressive mood and sleep disorders, with a total score of 76. The terminal automatically records the scores of each item and the total score, and makes a preliminary judgment on the emotional state based on the total score (≤6 points no depression, 7~16 points may be depressed, etc.), and transmits it to the system.

[0043] Step 5: Data Validation and Archiving The system verifies all collected data, checks the data range (e.g., the normal range of IL-17A is 0~15pg / mL), removes obviously abnormal data, marks the data collection time and the person who collected the data, and completes data archiving to provide standardized data for subsequent preprocessing.

[0044] Example 4: This embodiment further defines the structure of the data preprocessing module based on the above embodiments, such as... Figure 3 As shown, the data preprocessing module includes an outlier removal unit, a missing value imputation unit, a data standardization unit, a data alignment unit, and a detection error calibration unit;

[0045] The outlier removal unit uses the 3σ principle to remove outlier data. The normal ranges for serum inflammatory immune markers are as follows: neutrophil to lymphocyte ratio: 1-3; nitric oxide: 20-60 μmol / L; superoxide dismutase: 120-250 U / mL; interleukin-17A: 0-15 pg / mL; and interferon-α: 0-10 pg / mL. The missing value filling unit uses interpolation to fill in missing data; The data standardization unit uses the Z-score standardization method to eliminate dimensional differences in multimodal data; The data alignment unit achieves temporal alignment of multimodal data based on timestamps; The detection error calibration unit corrects systematic errors in serum inflammatory immune marker data.

[0046] The specific implementation process includes the following steps: Step 1: Outlier Removal (Executed by the Outlier Removal Unit) The 3σ principle is used to remove outliers from each modality of data: First, the mean μ and standard deviation σ of each index data are calculated, using the following formulas:

[0047] Where n is the number of samples, and xᵢ is the index value of the i-th sample.

[0048]

[0049] Unbiased standard deviation calculation is used to improve data accuracy; data exceeding the range of [μ-3σ, μ+3σ] are identified as outliers, with a focus on serum inflammatory immune marker data, and double verification is performed in combination with clinical normal ranges (such as NLR1~3) to avoid misjudgment; outlier data is recorded, the cause of the outlier is marked (such as testing equipment failure, poor patient cooperation), and transmitted to the storage interaction module for archiving.

[0050] Step 2: Missing value imputation (Missing value imputation unit execution) Linear interpolation was used to impute missing data: For continuous data such as serum inflammatory and immune markers and sleep monitoring, if the k-th data point is missing (denoted as x), then the missing data point is imputed. k Select two consecutive valid data points x before and after the missing data. k-1 (corresponding time t) k-1 ) and x k+1 (corresponding time t) k+1 The interpolation formula is:

[0051] This method ensures data continuity; for discrete data such as clinical scales, the mode imputation method is used, and the mode calculation formula is: (That is, select the score that appears most frequently as the imputation value); the imputed data is marked with the "imputation" label to avoid confusion with the original data.

[0052] Step 3: Detection error calibration (executed by the detection error calibration unit) The systematic errors of serum inflammatory immune marker data are calibrated, taking IL-17A as an example, using the following calibration formula:

[0053] Where k is taken as 1.0 (calibrated according to the blood testing equipment model), and Δx is the equipment systematic error (1 pg / mL). The error is random (following a normal distribution N(0,0.3²)); NLR is recalculated after calibrating neutrophil and lymphocyte counts. The neutrophil calibration formula is as follows:

[0054] Lymphocyte calibration formula:

[0055] NLR formula after calibration:

[0056] Ensure that the error of serum biomarker data after calibration is ≤3%, and transmit the calibrated data synchronously to the data standardization unit.

[0057] Step 4: Data Standardization (Data Standardization Unit Execution) The Z-score standardization method is used to transform all preprocessed multimodal data to the same order of magnitude (mean 0, standard deviation 1). The standardization formula is as follows:

[0058] Where x is the original value of the j-th indicator for the i-th patient, and μ is the mean of the j-th indicator ( ), σ is the standard deviation of the j-th indicator ( (σ≠0); standardization eliminates dimensional differences, ensuring that imaging data, serum biomarker data, and physiological signal data can be analyzed collaboratively.

[0059] Step 5: Data alignment (executed by the data alignment unit) Time alignment of multimodal data based on timestamps: Using the timestamp of the integrated monitoring terminal as the benchmark (denoted as t0), for cranial MRI data (once a month, timestamp t1), serum test data (once a week, timestamp t2), and clinical assessment data (once every two weeks, timestamp t3), the time alignment formula is as follows:

[0060] Where k=1,2,3, Δt is the acquisition interval of the integrated monitoring terminal, which is 1s; round(x) is a rounding function that aligns the timestamps of each modality data to t, ensuring that multimodal data at the same time point can be correlated, generating a standardized structured data set, and transmitting it to the multimodal fusion analysis module.

[0061] Example 5: This embodiment, based on the above embodiments, further defines the structure of the multimodal fusion analysis module, such as... Figure 4 As shown, the multimodal fusion analysis module includes a static feature extraction unit, a dynamic feature extraction unit, and a weighted fusion unit;

[0062] The static feature extraction unit is used to extract static features directly related to cognitive function and emotional state from the preprocessed dataset, including: imaging data features, serum inflammatory immune marker data features, clinical cognitive assessment data features, and clinical emotional assessment data features. The dynamic feature extraction unit is used to extract time-series dynamic features related to the changing trends of cognitive function and emotional state from the preprocessed dataset, including: neurophysiological signal data features and life behavior data features; The weighted fusion unit is used to perform weighted fusion of static and dynamic features, construct the multimodal fusion cognitive and emotional joint evaluation model based on the fused features, and generate a fusion feature vector through the model to provide input for subsequent cognitive function monitoring, emotional state assessment and risk warning.

[0063] The specific implementation process includes the following steps: Step 1: Feature extraction (executed by static feature extraction unit and dynamic feature extraction unit) (1) Static Feature Extraction: The static feature extraction unit receives the preprocessed structured data set and extracts static features through a CNN network, including: structural features such as the volume of high signal in white matter and the number of lacunar infarcts in imaging data; concentration features such as NLR, NO, and IL-17A in serum inflammatory immune marker data; and scale scoring features in clinical cognitive / emotional assessment data. The weights of serum inflammatory immune marker features are calculated using the following formula to ensure the weight proportion of key indicators:

[0064] In the formula, Let j be the feature weight of the j-th serum inflammatory immune marker. The absolute value of the Pearson correlation coefficient between the biomarker and the cognitive-emotional state is given, where M is the total number of serum biomarkers (M=5 in this example). According to this formula, the feature weights of IL-17A, NLR, and NO are set to 0.3, 0.3, and 0.2, respectively, while the combined weight of SOD and IFN-α is 0.2. Redundant features unrelated to cognitive and emotional states are then filtered out. (2) Dynamic feature extraction: The dynamic feature extraction unit extracts the temporal dynamic features (such as heart rate fluctuation trend and sleep cycle change) in the neurophysiological signal data (ECG, blood oxygen, sleep) through the LSTM network, captures the dynamic change pattern of the patient's cognitive and emotional state, and outputs the temporal dynamic feature vector.

[0065] Step 2: Feature weighted fusion (executed by the weighted fusion unit) (1) Feature selection: The correlation of static and dynamic features is checked, and redundant features with a correlation of less than 0.3 are removed (calculated by Pearson correlation coefficient) to ensure that all input features are highly correlated with cognitive function and emotional state; (2) Weighting: Feature weights were assigned according to preset rules, with serum inflammatory and immune marker data weighted at 25%, imaging data weighted at 35%, neurophysiological signal data weighted at 25%, lifestyle behavior data weighted at 10%, and clinical assessment data weighted at 5%. (3) Collaborative fusion: The attention weights of temporal dynamic features are calculated through the attention mechanism. The formulas for calculating the attention score and weights are as follows:

[0066] In the formula, The attention score for the i-th temporal dynamic feature is... Here is the attention weight matrix. For time-series dynamic feature vectors, For bias vectors, Here, T represents the total number of temporal features, and T is the attention weight. Static and dynamic features are weighted and fused based on the attention weights, using the following formula:

[0067] In the formula, For the final fused feature vector, =0.35 is the static feature weight coefficient. The static feature vector extracted by the CNN network; (4) Model deployment and processing: Input the fused feature vector into the trained multimodal fusion cognitive and emotional joint assessment model. The model outputs preliminary cognitive function and emotional state assessment reference results, which are transmitted to the cognitive function dynamic monitoring module and the emotional state assessment module together with the fused feature vector.

[0068] Step 3: Feature Validation and Optimization The system periodically (once a month) verifies the extracted features and fusion results, calculating the correlation between features and cognitive and emotional states using the Pearson correlation coefficient. The correlation calculation formula is as follows:

[0069] In the formula, Let be the correlation coefficient between the j-th feature and the cognitive-emotional joint state y. Let be the value of the j-th feature of the i-th patient. The mean of this feature. denoted as the mean of the cognitive-emotional joint state labels, and n is the number of patient samples. If the correlation is lower than the preset threshold (0.3), the feature weights are adjusted or features are re-screened, and the fusion algorithm parameters are optimized to ensure the accuracy and effectiveness of the fused feature vector.

[0070] The system periodically (once a month) verifies the extracted features and fusion results, calculates the correlation between features and cognitive and emotional states, and adjusts feature weights or re-selects features if the correlation is lower than a preset threshold, optimizes fusion algorithm parameters, and ensures the accuracy and effectiveness of the fused feature vector.

[0071] Example 6: This embodiment, based on the above embodiments, further defines the structure of the cognitive function dynamic monitoring module, such as... Figure 5 As shown, it includes a feature component extraction unit, a real-time evaluation unit, a dynamic curve drawing unit, an anomaly risk marking unit, and a time series tracking unit;

[0072] The feature component extraction unit is responsible for receiving the fusion feature vector generated by the multimodal fusion analysis module, accurately extracting the feature components in the vector corresponding to the cognitive function assessment dimensions (memory, attention, executive function, language function), synchronously associating the feature values ​​of the corresponding serum inflammatory immune markers, and transmitting them to the real-time assessment unit and the time sequence tracking unit. The real-time assessment unit pre-sets a cognitive function assessment index system, which includes four core assessment dimensions: memory, attention, executive function, and language function. Each dimension corresponds to a specific assessment threshold. Specifically, for the memory dimension, a score ≥3 on the MoCA scale's memory sub-item is considered normal; for the attention dimension, a score ≥3 on the MoCA scale's attention sub-item is considered normal; for the executive function dimension, a score ≥2 on the MoCA scale's executive function sub-item is considered normal; and for the language function dimension, a score ≥9 on the MMSE scale's language sub-item is considered normal. The real-time assessment unit receives feature components transmitted by the feature component extraction unit, compares the feature components of each dimension with the corresponding pre-set assessment thresholds, and combines this with the abnormality of serum inflammatory immune marker characteristic values ​​to generate a real-time cognitive function assessment result, which is divided into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal. Simultaneously, the assessment result and comparison data are transmitted to the dynamic curve plotting unit, the abnormal risk marking unit, and the storage and interaction module. The dynamic curve drawing unit receives the real-time cognitive function assessment results transmitted by the real-time assessment unit and the time-series fusion feature vector dataset transmitted by the time-series tracking unit. It draws the dynamic change curves of the four cognitive function assessment dimensions and the comprehensive cognitive function dynamic change curve at a frequency of once a day. The characteristic values ​​of serum inflammatory immune markers corresponding to each time node are marked in the curves, clearly presenting the changing trends of each dimension of cognitive function. The completed dynamic curves are transmitted to the terminal display module and the time-series tracking unit. The abnormal risk marking unit is connected to the feature component extraction unit and the real-time evaluation unit. It receives the feature components, evaluation results and serum inflammatory immune marker feature values ​​transmitted by the two units. When the interleukin-17A and neutrophil-to-lymphocyte ratio feature values ​​in the fused feature vector exceed the normal range, or the nitric oxide feature value is lower than the defined normal range, it automatically marks the cognitive function abnormality risk points of the corresponding evaluation dimension. It synchronously records the specific values ​​of the serum inflammatory immune markers corresponding to the abnormal features, the duration of the abnormality, and the change in the score of the corresponding cognitive dimension. It transmits the abnormal risk marking information to the time-series tracking unit, the risk warning module and the terminal display module. The time-series tracking unit is responsible for receiving the continuous time-series fusion feature vector transmitted by the feature component extraction unit, constructing a longitudinal monitoring dataset of the patient's cognitive function (retaining data for at least 30 days), and synchronizing the assessment results of the real-time assessment unit, the dynamic curve data of the dynamic curve drawing unit, and the abnormal labeling information of the abnormal risk labeling unit in real time to perform long-term dynamic tracking of various dimensions of cognitive function. When the fluctuation amplitude of a feature component of a certain assessment dimension is ≥10% for 3 consecutive days, or the fluctuation amplitude is ≥20% in a single instance, the dynamic curve drawing unit is automatically triggered to update the curve annotation, the abnormal risk labeling unit marks the abnormal fluctuation point, and feedback is synchronously sent to the real-time assessment unit to realize dynamic tracking of cognitive function.

[0073] The specific implementation process includes the following steps: Step 1: Feature Component Extraction (Executed by the Feature Component Extraction Unit) The system receives the fusion feature vector generated by the multimodal fusion analysis module, accurately extracts the feature components corresponding to the cognitive function assessment dimensions (memory, attention, executive function, and language function), and removes redundant features unrelated to cognitive function. The extracted feature components are then processed using the Z-score normalization method. The normalization formula is:

[0074] in These are the original values ​​of the cognitive dimension feature components. This is the mean of the characteristic component. This represents the unbiased standard deviation of the feature component. The feature values ​​of the corresponding serum inflammatory immune markers (IL-17A, NLR, NO) are simultaneously correlated, and the processed feature component is transmitted to the real-time evaluation unit and the time-series tracking unit.

[0075] Step 2: Real-time cognitive function assessment (executed by the real-time assessment unit) (1) A pre-set cognitive function assessment index system was established, with assessment thresholds for each dimension as follows: memory (MoCA memory sub-item ≥3 points), attention (MoCA attention sub-item ≥3 points), executive function (MoCA executive function sub-item ≥2 points), and language function (MMSE language sub-item ≥9 points). (2) Compare the extracted feature components of each dimension with the corresponding thresholds, and combine this with whether the feature values ​​of serum inflammatory immune markers are abnormal (referring to the normal range). Introduce a correlation verification formula between feature components and serum markers to improve the accuracy of the assessment:

[0076] In the formula The overall score is based on the cognitive dimension. =0.6 represents the weight of the cognitive feature components. =0.4 is the association weight of serum biomarkers (the weight allocation is in line with the association priority between cognitive characteristics and serum biomarkers in clinical practice). The values ​​of serum biomarkers are standardized (the standardization method is the same as in Example 4 and step 1 of this document); according to The score generates real-time cognitive function assessment results, which are divided into four levels: normal, mildly abnormal, moderately abnormal, and severely abnormal. Mild risk: Cognitive function assessment results are at the lower limit of the normal range, emotional state is possibly depressed, or IL-17A and NLR are mildly elevated (within 10% of the normal range) and NO is mildly decreased (within 10% of the normal range). The risk probability is 30%~50%, the warning signal is blue, and the intervention recommendation is regular monitoring (serum inflammatory immune marker test once a week, cognitive and emotional assessment once a month) and lifestyle adjustment (reasonable diet, regular exercise, smoking cessation and alcohol limitation). Moderate risk: Cognitive function assessment results show abnormalities in 1-2 dimensions, or the emotional state is moderately depressed, or IL-17A and NLR are moderately elevated (10%-30% above normal) and NO is moderately decreased (10%-30% below normal). The risk probability is 50%-70%, the warning signal is yellow, and the intervention recommendations are to seek medical attention in a timely manner, carry out targeted rehabilitation training (cognitive rehabilitation, emotional counseling), and have regular follow-up examinations (serum inflammatory immune markers test every 3 days, cognitive and emotional assessment every 2 weeks). Severe risk: Cognitive function assessment results show abnormalities in 3 or more dimensions, or the emotional state is severe depression, or IL-17A and NLR are severely elevated (more than 30% above the normal range) and NO is severely decreased (more than 30% below the normal range), with a risk probability ≥70%. The warning signal is red, and the intervention recommendations are immediate hospitalization, adjustment of the treatment plan (targeted suppression of inflammatory response and improvement of microcirculation), and strengthening of cognitive rehabilitation and emotional intervention.

[0077] (3) The assessment results, comparison data of each dimension, and serum biomarker association data are synchronously transmitted to the dynamic curve plotting unit, the abnormal risk marking unit, and the storage and interaction module.

[0078] Step 3: Understanding Dynamic Curve Drawing (Execution of the Dynamic Curve Drawing Unit) The system receives the evaluation results from the real-time evaluation unit and the time-series fusion feature vector dataset from the time-series tracking unit. It plots dynamic change curves for the four cognitive dimensions and a comprehensive cognitive function curve at a frequency of once per day. The curves are simultaneously labeled with the characteristic values ​​of serum inflammatory immune markers (such as IL-17A concentration and NLR value) corresponding to each time node, and normal / abnormal data are marked with different colors (green for normal and red for abnormal), clearly presenting the trend of cognitive function changes. The plotted dynamic curves are then transmitted to the terminal display module and the time-series tracking unit.

[0079] Step 4: Anomaly Risk Marking (Executed by the Anomaly Risk Marking Unit) Connected to the feature component extraction unit and the real-time evaluation unit, it receives relevant data and automatically marks cognitive function abnormality risk points when the following conditions occur: (1) The IL-17A and NLR characteristic values ​​exceed the defined normal range, or the NO characteristic value is below the normal range; (2) The fluctuation range of a certain cognitive dimension feature component is ≥10% (single time) or ≥20% (for 3 consecutive days); the formula for calculating the fluctuation range is:

[0080] In the formula For the fluctuation range of cognitive feature components, The feature component value at the current time. The previous moment's characteristic component value (the "previous moment" corresponds to the continuous monitoring node in the time series tracking); the cumulative calculation formula for the fluctuation amplitude over 3 consecutive days is: ; The system synchronously records the specific values ​​of serum biomarkers corresponding to abnormal risk points, the duration of abnormality, and the changes in cognitive dimension scores, and transmits the abnormal labeling information to the time-series tracking unit, the risk warning module, and the terminal display module.

[0081] Step 5: Cognitive Function Timing Tracking (Execution of Timing Tracking Unit) (1) Receive the continuous temporal fusion feature vector from the feature component extraction unit, construct a longitudinal monitoring dataset of patient cognitive function (retain at least 30 days of data), and synchronize the assessment results, dynamic curve data, and abnormal labeling information in real time; (2) Long-term dynamic tracking of each dimension of cognitive function. When the characteristic component of a certain dimension fluctuates by ≥10% for 3 consecutive days, or fluctuates by ≥20% in a single instance (the fluctuation amplitude is calculated according to the formula in step 4, which is completely consistent with the abnormal judgment criteria), the dynamic curve drawing unit is automatically triggered to update the curve label (add fluctuation abnormality mark), the abnormal risk mark unit marks the fluctuation abnormal point, and at the same time, it is fed back to the real-time assessment unit to recalculate the comprehensive assessment score of the cognitive dimension. (The calculation logic is the same as in step 2), reassess the cognitive function status to ensure the consistency of dynamic monitoring.

[0082] Example 7: This embodiment, based on the above embodiments, further defines the structure of the emotional state assessment module, such as... Figure 6 As shown, the emotional state assessment module includes an emotional feature extraction unit, an emotional level assessment unit, an emotional curve plotting unit, an emotional risk identification unit, and a connection analysis unit;

[0083] The emotional feature extraction unit is responsible for receiving the fusion feature vector generated by the multimodal fusion analysis module, accurately extracting the continuous temporal emotional feature components related to the emotional state in the vector, focusing on extracting the feature values ​​of serum inflammatory immune markers (interleukin-17A, neutrophil-to-lymphocyte ratio, nitric oxide) and the feature components corresponding to clinical emotional assessment data (HAMD-24 scale score), synchronously associating the relevant features of the corresponding cognitive function assessment, and transmitting them to the emotional level assessment unit, emotional risk identification unit and connection analysis unit; The emotional state assessment standard of the emotional level assessment unit adopts the Hamilton Depression Rating Scale (HAMD-24) scoring standard, combined with the correlation characteristics between serum inflammatory immune markers and emotional disorders (elevated interleukin-17A and neutrophil-to-lymphocyte ratio are positively correlated with emotional disorders, while elevated nitric oxide is negatively correlated with emotional disorders), and constructs an emotional state assessment sub-model; it receives continuous temporal emotional feature components transmitted by the emotional feature extraction unit, compares the HAMD-24 scale scoring features and serum inflammatory immune marker feature values ​​with the preset assessment standards within the emotional level assessment unit, and generates real-time emotional state assessment results, which are divided into four levels: no depression, possible depression, moderate depression, and severe depression. At the same time, the assessment results and comparison data are transmitted to the emotional curve plotting unit, emotional risk identification unit, storage interaction module, and connection analysis unit. The emotional curve drawing unit receives the real-time emotional state assessment results transmitted by the emotional level assessment unit and the continuous temporal emotional feature components transmitted by the emotional feature extraction unit. It draws the dynamic change curve of emotional state once a day. The curve is simultaneously marked with the HAMD-24 scale score and specific characteristic values ​​of serum inflammatory immune markers corresponding to each time node, clearly presenting the trend of emotional state change. The completed dynamic curve is transmitted to the terminal display module, the connection analysis unit and the storage interaction module. The emotional risk identification unit is connected to the emotional feature extraction unit and the emotional level assessment unit. It receives the feature components, assessment results and serum inflammatory immune marker characteristic values ​​transmitted by both units. Based on the defined normal range of serum inflammatory immune markers, it identifies the risk of emotional disorder. When the characteristic values ​​of interleukin-17A and neutrophil-to-lymphocyte ratio exceed the normal range, or the characteristic value of nitric oxide is lower than the normal range, and the corresponding emotional level assessment result is possible depression or above, the risk point of emotional disorder is automatically marked. The abnormal value of serum inflammatory immune markers, the duration of abnormality and the change in HAMD-24 scale score corresponding to the risk point are recorded simultaneously. The risk identification information is transmitted to the connection analysis unit, risk warning module and terminal display module. The connection analysis unit is connected to the emotion level assessment unit, the emotion curve plotting unit, and the cognitive function dynamic monitoring module, respectively. It receives the emotion state assessment results and emotion dynamic curve data, and synchronously acquires the cognitive function assessment results and cognitive dynamic curve data transmitted by the cognitive function dynamic monitoring module. It performs collaborative connection analysis on the change patterns of emotion state and cognitive function to capture the collaborative change characteristics of the two.

[0084] The specific implementation process includes the following steps: Step 1: Sentiment Feature Extraction (Executed by the Sentiment Feature Extraction Unit) The system receives the fusion feature vector generated by the multimodal fusion analysis module, accurately extracts continuous temporal emotional feature components related to emotional state, focuses on extracting feature values ​​of serum inflammatory immune markers (IL-17A, NLR, NO) and feature components corresponding to HAMD-24 scale scores, and removes irrelevant and redundant features. The extracted feature components are then standardized using Z-score, with the standardization formula as follows:

[0085] These are the original values ​​of the sentiment feature components. This is the mean of the characteristic component. This represents the unbiased standard deviation of the feature component. It is then synchronously correlated with relevant features of the corresponding cognitive function assessment and transmitted to the emotion level assessment unit, emotion risk identification unit, and connectivity analysis unit.

[0086] Step 2: Emotional State Level Assessment (Executed by the Emotional Level Assessment Unit) (1) Pre-set emotional state assessment criteria, adopt the HAMD-24 scale scoring criteria, and combine the correlation characteristics between serum inflammatory immune markers and emotional disorders to construct an emotional state assessment sub-model; (2) Receive the feature components of the emotion feature extraction unit, compare the HAMD-24 scale scoring features and serum inflammatory immune marker feature values ​​with preset standards, and introduce a serum marker correlation verification formula to improve the accuracy of emotion level assessment:

[0087] In the formula For the comprehensive emotional assessment score, =0.55 is the weight of the HAMD-24 scale score. =0.45 is the association weight of serum markers (which is consistent with the strong association between serum inflammation and immunity and affective disorders in clinical practice). The values ​​are the standardized scores of the scale. The values ​​are the standardized values ​​of serum biomarkers; based on The scores are mapped to HAMD-24 scale rating intervals to generate real-time emotional state assessment results, which are divided into four levels: no depression (≤6 points), possible depression (7~16 points), moderate depression (17~23 points), and severe depression (≥24 points). (3) The evaluation results and comparison data are synchronously transmitted to the emotional curve drawing unit, the emotional risk identification unit, the storage interaction module and the connection analysis unit.

[0088] Step 3: Drawing the emotional dynamic curve (executed in the emotional curve drawing unit) The system receives the assessment results from the emotional level assessment unit and the continuous temporal emotional feature components from the emotional feature extraction unit, and plots a dynamic change curve of emotional state once a day. The curve is simultaneously labeled with the total score of the HAMD-24 scale, the score of each sub-item, and the specific characteristic values ​​of serum inflammatory immune markers at each time point. Different colors are used to distinguish the emotional level (green: no depression, yellow: possible depression, orange: moderate depression, red: severe depression), clearly presenting the trend of emotional state change. The completed dynamic curve is transmitted to the terminal display module, connected to the analysis unit, and the storage and interaction module.

[0089] Step 4: Identification of Emotional Disorder Risks (Executed by the Emotional Risk Identification Unit) Connected to the emotion feature extraction unit and the emotion level assessment unit, it receives relevant data and, referring to the normal range of serum inflammatory immune markers as defined in claim 4, identifies the risk of developing emotional disorders: A risk association determination formula is introduced to quantify the degree of association between abnormal serum markers and emotional abnormalities.

[0090] In the formula This represents the risk association value for affective disorders. , , The feature weights for the three components are 0.3, 0.3, and 0.2, respectively. The values ​​after calibration for each marker. , These are the mean and unbiased standard deviation of each marker, respectively; when When the threshold is ≥0.5 (preset threshold) and the emotional level assessment result is possibly depressed or above, the risk point of emotional disorder is automatically marked; the abnormal value of serum markers, the duration of abnormality, and the change in HAMD-24 scale score corresponding to the risk point are recorded simultaneously, and the risk identification information is transmitted to the connection analysis unit, risk warning module and terminal display module.

[0091] Step 5: Cognitive-Affective Connectivity Analysis (Executed by the Connectivity Analysis Unit) It connects to the emotion level assessment unit, the emotion curve plotting unit, and the cognitive function dynamic monitoring module respectively, receiving emotion state assessment results and emotion dynamic curve data, and simultaneously acquiring cognitive function assessment results and cognitive dynamic curve data; it performs co-connection analysis on the changing patterns of both, and calculates the cognitive-emotion correlation using the Pearson correlation coefficient, with the formula as follows:

[0092] In the formula , These represent the cognitive and emotional comprehensive assessment scores of the i-th patient. , , where n is the mean of the two values, and n is the number of patient samples. When the correlation is ≥0.6 and both cognitive and emotional abnormalities occur simultaneously, comorbidity characteristics are captured: when an abnormal worsening of emotional state is detected (e.g., from moderate depression to severe depression) accompanied by a decline in cognitive function (e.g., attention dimension changes from normal to mildly abnormal), or when abnormal serum inflammatory immune markers (e.g., elevated IL-17A) affect both simultaneously, a connectivity analysis suggestion is generated and synchronously fed back to the emotional level assessment unit, the cognitive function dynamic monitoring module, and the risk warning module to provide a basis for comorbidity warning.

[0093] Example 8: This embodiment, based on the above embodiments, further defines the structure of the risk warning module, such as... Figure 7 As shown, the risk warning module includes a risk feature receiving unit, a risk probability calculation unit, a warning threshold comparison unit, a warning signal generation unit, and an intervention suggestion generation unit;

[0094] The risk feature receiving unit is responsible for synchronously receiving the cognitive function dynamic status (including cognitive function assessment results, cognitive dynamic change curve data, and abnormal risk marker information) transmitted by the cognitive function dynamic monitoring module, the emotional state assessment results (including emotional level, emotional dynamic curve data, and emotional disorder risk point information) transmitted by the emotional state assessment module, and the fusion feature vector transmitted by the multimodal fusion analysis module, and synchronously transmitting them to the risk probability calculation unit. The risk probability calculation unit has a preset risk warning model. Combining the correlation characteristics between serum inflammatory immune markers and cognitive impairment and emotional disorders in CSVD, it calculates the initial risk probability of cognitive impairment and emotional disorders in patients based on all data transmitted by the risk feature receiving unit. During the calculation process, it strictly follows the risk probability calculation formula defined in the invention and introduces patient individual difference correction coefficients (age correction coefficient, disease duration correction coefficient) to dynamically correct the initial risk probability and obtain the final risk probability. At the same time, it calculates the risk contribution of each serum inflammatory immune marker to clarify the degree of influence of abnormality of each marker on the risk probability, and transmits the final risk probability and risk contribution data to the warning threshold comparison unit. The warning threshold comparison unit presets three warning thresholds (corresponding to mild risk, moderate risk, and severe risk). It receives the final risk probability transmitted by the risk probability calculation unit, compares it with the three warning thresholds one by one, and determines the risk level (mild, moderate, and severe). At the same time, it combines the dynamic abnormality of cognitive function, the level of emotional state assessment, and the degree of abnormality of serum inflammatory immune markers to verify the rationality of the risk level determination. The determined risk level and the comparison result are then transmitted to the warning signal generation unit. The warning signal generation unit is connected to the warning threshold comparison unit. Based on the determined risk level, it generates a warning signal of the corresponding level and clarifies the warning type (cognitive impairment warning, emotional disorder warning, and comorbidity warning). The warning signal includes a warning level indicator (blue for mild, yellow for moderate, and red for severe), risk level, final risk probability, and information on the serum inflammatory immune marker with the highest risk contribution. It is also synchronously associated with the cognitive function abnormality dimension and the emotional state abnormality level. The generated warning signal is transmitted in real time to the warning suggestion generation unit, the terminal display module, and the storage interaction module, and simultaneously triggers the cognitive function dynamic monitoring module and the emotional state assessment module to update the abnormality marker information synchronously. The intervention suggestion generation unit is connected to the early warning threshold comparison unit and the risk probability calculation unit. Based on the determined risk level, final risk probability, abnormality of serum inflammatory immune markers (abnormality type and degree), and abnormality of cognitive function and emotional state, it generates targeted intervention suggestions.

[0095] The specific implementation process includes the following steps: Step 1: Risk Feature Reception (Executed by the Risk Feature Reception Unit) The system synchronously receives the cognitive function dynamic status (assessment results, dynamic curve data, and abnormality marker information) transmitted by the cognitive function dynamic monitoring module, the emotional state assessment results (emotional level, dynamic curve data, and risk point information) transmitted by the emotional state assessment module, and the fusion feature vector transmitted by the multimodal fusion analysis module. It focuses on extracting the feature values ​​of serum inflammatory immune markers (IL-17A, NLR, NO) from the fusion feature vector, sorts and removes all data, eliminates irrelevant and redundant information, ensures data integrity and consistency, and synchronously transmits them to the risk probability calculation unit.

[0096] Step 2: Risk Probability Calculation (Executed by the Risk Probability Calculation Unit) (1) The preset risk warning model is invoked, and the correlation characteristics between serum inflammatory immune markers and cognitive and emotional disorders of CSVD are combined. Based on all received data, the initial risk probability formula is calculated as follows:

[0097] The formula for the comprehensive value of serum biomarker risk characteristics is as follows:

[0098] in, (x) is the Sigmoid activation function ( W1, W2, W3, and W4 are the weight matrices for fusion features, cognitive features, emotional features, and serum biomarker features, respectively. To fuse feature vectors, For cognitive function feature vectors, For emotional state feature vectors, This is a composite value of risk characteristics of serum biomarkers; For model bias terms; , , The values ​​for each marker are their respective weights, and x represents the calibrated values ​​for each marker. (2) Introduce an individual difference correction coefficient to dynamically correct the initial risk probability. The correction formula is as follows:

[0099] Formula for calculating relative age:

[0100] P represents the final risk probability after correction, and P represents the initial risk probability. =0.005 (age correction factor) =0.02 (disease duration correction factor); The values ​​are relative to age (with 60 years old as the benchmark for high incidence of CSVD). The patient's actual age (in years). The duration of the patient's illness (in years); (3) Calculate the risk contribution of each serum inflammatory immune marker, using the following formula:

[0101] in, Risk contribution (%) to the j-th serum biomarker; Let be the weighting coefficient of the j-th serum biomarker. The calibrated value is given; M is the total number of serum biomarkers (M=5 in this example); this formula clarifies the degree of influence of abnormalities in each biomarker on the risk probability. (4) Transmit the final risk probability and risk contribution data to the early warning threshold comparison unit.

[0102] Step 3: Early warning threshold comparison (executed by the early warning threshold comparison unit) (1) Preset three-level warning thresholds: mild risk (30%~50%), moderate risk (50%~70%), and severe risk (≥70%). The threshold standards refer to the normal range of serum markers and the preferred scheme of the invention. (2) Receive the final risk probability and compare it with the three-level warning threshold one by one to determine the risk level; at the same time, combine the dynamic abnormality of cognitive function (such as the number of abnormal dimensions), the level of emotional state assessment, and the degree of abnormality of serum markers (mild / moderate / severe abnormality) to verify the rationality of the risk level determination and avoid misjudgment. (4) The determined risk level and comparison results are transmitted to the early warning signal generation unit.

[0103] Step 4: Early warning signal generation (executed by the early warning signal generation unit) Connected to the early warning threshold comparison unit, it generates corresponding early warning signals based on the risk level, clarifying the early warning type (cognitive impairment early warning, emotional disorder early warning, and comorbidity of both). The early warning signal includes: early warning level identifier (blue = mild, yellow = moderate, red = severe), risk level, final risk probability, and serum biomarker information with the highest risk contribution, and is synchronously associated with cognitive abnormality dimension and emotional abnormality level. The generated early warning signal is transmitted in real time to the intervention suggestion generation unit, terminal display module, and storage interaction module, and at the same time triggers the cognitive function dynamic monitoring module and emotional state assessment module to synchronously update the abnormal labeling information.

[0104] Step 5: Intervention Recommendation Generation (Executed by the Intervention Recommendation Generation Unit) Connected to the early warning threshold comparison unit and the risk probability calculation unit, it generates targeted intervention suggestions based on the risk level, final risk probability, abnormal serum biomarkers, and cognitive / emotional abnormalities, aligning with the invention's three-level early warning mechanism. (1) Mild risk: Regular monitoring (serum biomarker testing once a week, cognitive / emotional assessment once a month), lifestyle adjustment (reasonable diet, regular exercise, smoking cessation and alcohol limitation); (2) Moderate risk: timely medical treatment, rehabilitation training (cognitive rehabilitation, emotional counseling), and regular follow-up examinations (serum marker testing every 3 days and assessment every 2 weeks); (3) Severe risk: Immediate hospitalization, adjustment of treatment plan (suppressing inflammatory response, improving microcirculation), and strengthening rehabilitation and emotional intervention; Intervention recommendations are linked to early warning signals and transmitted synchronously to the terminal display module and storage interaction module to provide accurate references for clinical intervention.

[0105] Example 9:

[0106] Based on the above embodiments, this embodiment further defines the structure of the storage interaction module. The storage interaction module adopts a distributed storage architecture, uses the AES encryption algorithm to ensure data security, and supports cloud backup function. The data interaction interface adopts the HL7 protocol to achieve seamless integration with hospital-related systems. The storage interaction module adopts a distributed storage architecture with a storage capacity of no less than 10TB. It stores raw data collected by the multimodal data acquisition module, preprocessed data sets generated by the data preprocessing module, fusion feature vectors and parameters of the multimodal fusion cognitive and emotional joint assessment model generated by the multimodal fusion analysis module, various assessment results, dynamic curve data and anomaly marker information generated by the cognitive function dynamic monitoring module and the emotional state assessment module, and warning records, warning signals, and intervention suggestions generated by the risk warning module. Simultaneously, a dedicated database of serum inflammatory and immune biomarkers is established, linking patient cognitive function, emotional state assessment results, and warning information to form a longitudinal patient monitoring dataset. This dataset supports rapid retrieval by detection time, biomarker type, and patient ID, facilitating subsequent model optimization and reuse in clinical research. The storage interaction module uses the AES-256 encryption algorithm to encrypt patient privacy data, complying with the requirements of the "Medical Data Security Guidelines." It provides dual security protection through encrypted storage and encrypted transmission to prevent patient privacy leaks. It also supports cloud backup, with a preset automatic backup mechanism every morning to back up all stored data to a dedicated cloud server, ensuring data security. Data is retained for at least 3 years without loss. The storage interaction module provides a standardized data interaction interface, designed using the HL7 protocol, enabling seamless integration with hospital electronic medical record systems and clinical diagnosis and treatment systems. It supports bidirectional synchronous data updates, allowing medical staff to quickly access patient monitoring data, assessment results, and longitudinal changes in serum inflammatory and immune markers through the hospital system. Clinical diagnosis and treatment information can also be synchronized to this system for data sharing. Furthermore, the storage interaction module supports real-time data writing and rapid querying, with a query response time ≤1 second. It also features a data log recording function, detailing all data writing, querying, modification, and deletion operations for easy traceability and management. The cognitive function dynamic monitoring module, emotional state assessment module, and risk warning module are connected. When abnormal changes in cognitive function or emotional state are detected, or when the risk probability reaches the warning threshold, the risk warning module triggers a warning signal. The cognitive function dynamic monitoring module and emotional state assessment module jointly generate an abnormal change analysis report, clarifying the specific dimensions of cognitive function and emotional state decline and the abnormal impact of serum inflammatory and immune markers. This report is synchronously stored in the storage interaction module and archived in association with the corresponding patient's longitudinal monitoring dataset.

[0107] The specific implementation process includes the following steps: Step 1: Storage Architecture Setup A distributed storage architecture is adopted to build a storage system with a storage capacity of 10TB (expandable). The system is divided into multiple storage partitions: raw data area (stores raw data collected from multiple modalities), preprocessed data area (stores standardized datasets), fusion feature area (stores fusion feature vectors and model parameters), evaluation and early warning area (stores cognitive / emotional evaluation results and early warning records), and dedicated database area (dedicated database for serum inflammatory immune biomarkers). This ensures that data is stored in a categorized manner and is easy to retrieve.

[0108] Step 2: Data Storage and Encryption (1) Receive data (raw data, preprocessed data, evaluation results, early warning signals, etc.) transmitted by each module and write it to the corresponding storage partition in real time. Real-time data writing is supported and the query response time is ≤1s. (2) The AES-256 encryption algorithm is used to encrypt patient privacy data (such as serum test data and personal information) to achieve dual protection of encrypted storage and encrypted transmission and prevent privacy leakage; (3) Establish a dedicated database of serum inflammatory immune markers, linking patient ID, cognitive function assessment results, emotional state assessment results, and early warning information to form a longitudinal monitoring dataset for patients, supporting rapid retrieval by detection time, marker type, and patient ID.

[0109] Step 3: Data Backup The system is set to automatically back up all stored data to a dedicated cloud server at 2:00 AM every day, and the backup data will be retained for at least 3 years. Manual backup is also supported, and medical staff can back up important data (such as early warning records and longitudinal monitoring data) in a timely manner as needed to ensure that the data is not lost. The backup data is protected by the same encryption method.

[0110] Step 4: Data Interaction and Sharing (1) The HL7 protocol is used to design a standardized data interaction interface to achieve seamless connection with the hospital's electronic medical record system and clinical diagnosis and treatment system, and to support bidirectional synchronous data updates; (2) Medical staff can quickly access patients’ monitoring data, cognitive / emotional assessment results, longitudinal changes in serum markers, and early warning records through the hospital system; at the same time, they can synchronize clinical diagnosis and treatment information (such as medication and treatment plans) to this system to achieve data sharing; (3) It has a data log recording function, which records all data writing, querying, modification and deletion operations in detail, and marks the operator and operation time, which facilitates traceability management and complies with medical data management standards.

[0111] Step 5: Data Connection Archiving When abnormal changes occur in cognitive function or emotional state, or when the risk probability reaches the warning threshold, the system receives an abnormal change analysis report jointly generated by the cognitive function dynamic monitoring module and the emotional state assessment module. The report is then linked and archived with the corresponding patient's longitudinal monitoring dataset to facilitate subsequent clinical review and model optimization.

[0112] Example 10: This embodiment, based on the above embodiments, further defines the structure of the terminal display module, such as... Figure 8 As shown, the terminal display module includes five functional units: a data receiving unit, a visualization rendering unit, a multi-terminal adaptation unit, an interactive operation unit, and an information push unit. The data receiving unit is responsible for synchronously receiving the cognitive function monitoring results and cognitive dynamic change curve data transmitted by the cognitive function dynamic monitoring module, the emotional state assessment results and emotional dynamic curve data transmitted by the emotional state assessment module, the graded warning signals and intervention suggestions transmitted by the risk warning module, and the serum inflammatory immune marker detection data and longitudinal change trend data transmitted by the storage and interaction module, and synchronously transmitting them to the visualization rendering unit. The visualization rendering unit is responsible for converting the received data into an intuitive visualization format. Specific rendering content includes: dynamic change curves of various dimensions of cognitive function (labeling the characteristic values ​​of serum inflammatory and immune markers at corresponding time points), dynamic change curves of emotional state (labeling HAMD-24 scale scores and corresponding serum marker values), warning signal level indicators (blue for mild, yellow for moderate, and red for severe), details of cognitive / emotional assessment results (including scores for each dimension and the basis for level determination), a list of intervention recommendations (sorted by priority, specifying the frequency of serum marker monitoring and assessment), and a longitudinal change chart of serum inflammatory and immune markers. During the rendering process, clear color differentiation is used (green for normal data, red for abnormal data, and the corresponding warning level color for warning data). Key nodes are labeled on curve data to facilitate quick identification of core information by medical staff and patients. The rendered visualization content is then transmitted to the multi-terminal adaptation unit. The multi-terminal adaptation unit is responsible for adapting the display content output by the visualization rendering unit to the display requirements of different terminals, supporting synchronous display on doctor workstations, patient mobile devices, and nursing terminals, and optimizing the display layout according to terminal type: the PC terminal displays complete and detailed data (including all monitoring indicators, curves, early warning records, and raw data of serum biomarkers); the mobile APP displays simplified content (including core assessment results, early warning prompts, simplified curves, and intervention suggestions); and the tablet terminal displays key monitoring content (including real-time early warning signals, core indicators of cognitive / emotional status, and abnormal serum biomarker prompts), ensuring the adaptability and readability of the content displayed on different terminals, and transmitting the adapted content to the interactive operation unit and information push unit; The interactive operation unit is connected to the multi-terminal adaptation unit and the storage interaction module, supporting targeted interactive operations by medical staff and patients. Specifically, medical staff can manually adjust the warning threshold display parameters, export cognitive / emotional assessment reports, and query longitudinal monitoring data of patients' serum inflammatory immune markers and historical warning records through the PC terminal; patients can view their own assessment results and intervention suggestions through the mobile APP, and manually upload lifestyle data; nursing staff can mark the warning signal processing status and view patients' real-time monitoring data through the tablet terminal. All interactive operation records are synchronously transmitted to the storage interaction module for archiving, facilitating traceability and management. The information push unit is connected to the risk warning module and the multi-terminal adaptation unit. When the risk warning module triggers a graded warning signal, it automatically pushes the warning signal, corresponding intervention suggestions, and relevant monitoring data (with a focus on abnormal serum inflammatory and immune marker information) to the corresponding terminals: pushing complete warning details and patient longitudinal monitoring data to the doctor's workstation, pushing real-time warning prompts and the patient's current monitoring status to the nursing terminal, and pushing simplified warning prompts and actionable intervention suggestions to the patient's mobile terminal; it also supports the regular push of cognitive / emotional assessment results summaries and serum inflammatory and immune marker detection reminders, ensuring that relevant personnel can promptly obtain the monitoring data, assessment results, warning signals, and intervention suggestions transmitted by each module, transforming them into an intuitive visual form, while adapting to the display needs of different terminals (doctor's workstation, patient's mobile terminal, and nursing terminal). Through interactive operation and information push, it ensures that medical staff and patients can promptly obtain core information, taking into account both clinical diagnosis and treatment and patient self-monitoring needs, and achieving practical and convenient information display.

[0113] The specific implementation process includes the following steps: Step 1: Data Reception and Processing (Executed by the Data Reception Unit) The system synchronously receives cognitive monitoring results and dynamic curve data transmitted from the cognitive function dynamic monitoring module, emotional assessment results and dynamic curve data transmitted from the emotional state assessment module, graded early warning signals and intervention suggestions transmitted from the risk warning module, and serum inflammatory immune marker detection data and longitudinal trend data transmitted from the storage and interaction module. All received data is classified and organized, redundant information is removed, and content is filtered and displayed according to terminal type to ensure data integrity and timeliness, and synchronously transmitted to the visualization rendering unit.

[0114] Step 2: Visual Rendering (Execution of the Visual Rendering Unit) As the core display unit, it transforms the received data into an intuitive visual form, specifically rendering the following content: (1) Curve type: dynamic curves of various dimensions of cognitive function (with characteristic values ​​of serum markers), dynamic curves of emotional state (with HAMD-24 scores and serum marker values), and longitudinal change charts of serum inflammatory and immune markers; (2) Identification type: warning signal level indicators (blue / yellow / red), cognitive / emotional assessment level indicators; (3) Details: Cognitive / emotional assessment results (scores for each dimension, criteria for grade determination), list of intervention recommendations (sorted by priority), specific test values ​​of serum markers and abnormality warnings; During the rendering process, color differentiation is used: green for normal data, red for abnormal data, and color corresponding to the warning level for warning data. Key nodes (abnormal points, warning points) are marked on the curves to facilitate quick identification of core information. The rendered visualization content is transmitted to the multi-terminal adaptation unit.

[0115] Step 3: Multi-terminal adaptation (executed by the multi-terminal adaptation unit) Adapt the visual rendering content to the display requirements of different terminals, support simultaneous display on multiple terminals, and optimize the layout: (1) Line, early warning record, raw data of serum markers, abnormal analysis report, supporting data export and parameter adjustment; (2) Patient mobile terminal (mobile APP): Displays simplified content, including core assessment results, early warning prompts, simplified curves, actionable intervention suggestions, and supports uploading of lifestyle behavior data; (3) Nursing terminal (tablet): Displays key monitoring content, including real-time early warning signals, cognitive / emotional core indicators, and abnormal serum markers. It supports status marking of early warning signal processing. The adapted content is transmitted to the interactive operation unit and information push unit.

[0116] Step 4: Interactive Operation (Execution of the Interactive Operation Unit) Connects to multi-terminal adaptation units and storage interaction modules to support targeted interactive operations: (1) Medical staff (PC): Manually adjust the display parameters of the warning threshold, export the cognitive / emotional assessment report, query the longitudinal data of patient serum markers and historical warning records, and archive the operation records synchronously; (2) Patients (mobile APP): View their own assessment results and intervention suggestions, and manually upload lifestyle data such as diet and exercise; (3) Nursing staff (tablet): Mark the status of warning signal processing (unprocessed / processed), view real-time patient monitoring data, and transmit operation records synchronously to the storage and interaction module for archiving.

[0117] Step 5: Information Push (Executed by the Information Push Unit) It connects with the risk warning module and multi-terminal adaptation unit to achieve accurate information push: (1) Early warning push: When the risk warning module triggers an early warning signal, it automatically pushes the early warning signal, intervention suggestions, and abnormal serum marker information to the corresponding terminal. The doctor pushes the complete details, the nurse pushes the real-time prompts, and the patient pushes the simplified prompts. (2) Regular push notifications: Regularly (once a week) push notifications of cognitive / emotional assessment results and serum inflammatory immune marker detection reminders to ensure that relevant personnel can obtain the information in a timely manner and improve monitoring compliance.

[0118] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function, characterized in that, It includes a multimodal data acquisition module, a data preprocessing module, a multimodal fusion analysis module, a cognitive function dynamic monitoring module, an emotional state assessment module, a risk warning module, a storage and interaction module, and a terminal display module; The multimodal data acquisition module collects raw data from various dimensions of CSVD patients; The data preprocessing module cleans, standardizes, reduces noise, aligns, and calibrates the detection error of the collected multi-dimensional raw data to generate a preprocessed dataset. The multimodal fusion analysis module performs multimodal fusion analysis on the preprocessed dataset, extracts cognitive function and emotional state correlation features from each modality, constructs a multimodal fusion cognitive and emotional joint evaluation model, and generates a fusion feature vector; The cognitive function dynamic monitoring module assesses the cognitive function status of CSVD patients in real time based on fused feature vectors, plots the dynamic change curve of cognitive function, and realizes dynamic tracking of cognitive function status. The emotional state assessment module assesses the emotional state of CSVD patients in real time based on fused feature vectors, identifies the risk of emotional disorder, plots dynamic changes in emotional state, and obtains emotional state assessment results. The risk warning module calculates the risk probability of cognitive impairment and emotional disorders based on the dynamic status of cognitive function, the results of emotional state assessment, and the fusion feature vector, and generates graded warning signals and intervention suggestions based on the three-level warning threshold. The storage interaction module stores relevant data, provides a data interaction interface, and enables connection with the hospital's electronic medical record system and clinical diagnosis and treatment system. The terminal display module presents monitoring results, emotional state assessment results, early warning signals, and intervention suggestions in a visual format.

2. The multimodal fusion-based CSVD cognitive function dynamic monitoring and early warning system according to claim 1, characterized in that, The multimodal data acquisition module includes an integrated monitoring terminal and an external acquisition device; The integrated monitoring terminal has built-in electrocardiogram sensor, blood oxygen sensor, and sleep monitoring sensor to collect neurophysiological signal data; The electrocardiogram sensor collects electrocardiogram signal data as neurophysiological signal data; The blood oxygen sensor collects blood oxygen saturation data; The sleep monitoring sensor collects electroencephalogram (EEG) signal data and sleep cycle data; The external data acquisition devices include a cranial MRI machine, a blood testing machine, a cognitive assessment terminal, and an emotion assessment terminal; The cranial MRI equipment acquires imaging data; The blood testing equipment collects serum inflammatory immune marker data; The cognitive assessment terminal collects clinical cognitive assessment data; The emotion assessment terminal collects clinical emotion assessment data.

3. The multimodal fusion-based CSVD cognitive function dynamic monitoring and early warning system according to claim 2, characterized in that, The imaging data acquired by the cranial MRI equipment includes the volume of high signal in the white matter, the number of lacunar infarcts, and the number and location of cerebral microbleeds. The serum inflammatory immune marker data collected by the blood testing device include the neutrophil-to-lymphocyte ratio, nitric oxide, superoxide dismutase, interleukin-17A, and interferon-α concentration data. The clinical cognitive assessment data collected by the cognitive assessment terminal includes the scores of the MMSE scale and the MoCA scale. For those with less than 12 years of education, additional points are added to their total MoCA score. The clinical emotion assessment data collected by the emotion assessment terminal includes HAMD-24 scale score data.

4. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function according to any one of claims 1 to 3, characterized in that, The data preprocessing module includes an outlier removal unit, a missing value imputation unit, a data standardization unit, a data alignment unit, and a detection error calibration unit; The outlier removal unit uses the 3σ principle to remove outlier data. The missing value filling unit uses interpolation to fill in missing data; The data standardization unit uses the Z-score standardization method to eliminate dimensional differences in multimodal data; The data alignment unit achieves temporal alignment of multimodal data based on timestamps; The detection error calibration unit corrects systematic errors in serum inflammatory immune marker data.

5. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function according to claim 1 or 2, characterized in that, The multimodal fusion analysis module includes a static feature extraction unit, a dynamic feature extraction unit, and a weighted fusion unit; The static feature extraction unit is used to extract static features directly related to cognitive function and emotional state from the preprocessed dataset. The dynamic feature extraction unit is used to extract time-series dynamic features related to the changing trends of cognitive function and emotional state from the preprocessed dataset; The weighted fusion unit is used to perform weighted fusion of static and dynamic features, construct the multimodal fusion cognitive and emotional joint evaluation model based on the fused features, and generate a fusion feature vector through the model to provide input for subsequent cognitive function monitoring, emotional state assessment and risk warning.

6. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function according to claim 1 or 2, characterized in that, The cognitive function dynamic monitoring module includes a feature component extraction unit, a real-time evaluation unit, a dynamic curve plotting unit, an abnormal risk marking unit, and a time series tracking unit.

7. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function according to claim 1 or 2, characterized in that, The emotional state assessment module includes an emotional feature extraction unit, an emotional level assessment unit, an emotional curve plotting unit, an emotional risk identification unit, and a connection analysis unit.

8. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function according to claim 1 or 2, characterized in that, The risk warning module includes a risk feature receiving unit, a risk probability calculation unit, a warning threshold comparison unit, a warning signal generation unit, and an intervention suggestion generation unit.

9. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function according to claim 1 or 2, characterized in that, The storage interaction module adopts a distributed storage architecture, uses AES encryption algorithm to ensure data security, and supports cloud backup function; the data interaction interface adopts HL7 protocol to achieve seamless integration with hospital-related systems.

10. A multimodal fusion-based dynamic monitoring and early warning system for CSVD cognitive function according to claim 1 or 2, characterized in that, The terminal display module includes a data receiving unit, a visualization rendering unit, a multi-terminal adaptation unit, an interactive operation unit, and an information push unit.

Images