Method and system for comprehensively monitoring daytime and night symptoms of Parkinson's disease

Abstract

The invention discloses a method and a system for comprehensively monitoring daytime and nighttime symptoms of Parkinson's disease, which are used for solving the problems of fragmentization, strong subjectivity, low calculation efficiency and the like of the existing monitoring. The system synchronously collects real-time clock, body position, EOG, EMG and IMU (inertial measurement unit) multi-mode signals, hierarchically identifies daytime activity, night sleep, pre-sleep and initial awakening stages, and then adaptively activates corresponding modules to analyze symptoms: monitoring movement symptoms and tumble in daytime and attributing, identifying sleep disorder at night, and identifying sleep disorder at night. A morning stiffness severity index is generated by fusing time, myoelectricity and motion signals in the initial awakening period; the characteristics are mapped into UPDRS / PDDS scores based on machine learning, a structured report containing visualization and historical comparison is generated, a diagnosis and treatment closed loop of'monitoring-analysis-reporting-doctor evaluation-treatment adjustment 'is constructed, objective continuous monitoring is achieved, Parkinson's disease diagnosis and treatment are assisted, and medical efficiency is improved.

Description

Technical Field

[0001] This invention relates to the field of neurodegenerative disease monitoring and clinical auxiliary diagnosis technology, and more specifically, to a method and system for monitoring Parkinson's disease (PD) symptoms based on multimodal wearable sensing, hierarchical state recognition and adaptive intelligent analysis. Background Technology

[0002] Parkinson's disease is the second most prevalent neurodegenerative disease worldwide, affecting over 10 million people globally, with a prevalence rate of 1.37% in people over 60 years of age in my country. The core challenges in its clinical management lie in the volatility of symptoms (such as end-of-dose symptoms and dyskinesia) and their environmental dependence (differences between home and laboratory findings): traditional diagnosis relies on short-term in-hospital motor function tests (such as physician observation) and subjective scales (such as the MDS-UPDRS, the Unified Parkinson's Disease Rating Scale), while a single outpatient assessment is insufficient to reflect the patient's true symptoms in daily life; patients' home-based paper / electronic diaries suffer from recall bias and cannot capture transient symptoms (such as frozen gait or nocturnal limb tics); simultaneously, symptom quantification is disconnected from medication timing, making it difficult for physicians to establish a dynamic "blood drug concentration-motor symptom" model, and medication adjustments rely heavily on experience, thus limiting treatment efficiency.

[0003] To compensate for the shortcomings of traditional evaluation methods, existing technologies have seen the emergence of technologies such as... Watches Wearable devices such as belts are available, but these technologies still have significant limitations and cannot meet the clinical needs for comprehensive and accurate monitoring of Parkinson's disease.

[0004] The core shortcomings of existing Parkinson's disease symptom monitoring technologies can be summarized in the following seven points:

[0005] Fragmented monitoring lacks a continuous 24-hour perspective: Existing protocols either focus only on daytime motor symptoms (such as tremor and gait) or only analyze nighttime sleep structure, failing to cover the entire cycle of "pre-sleep - nighttime sleep - first awakening - daytime activity". This makes it difficult to reveal the intrinsic relationship between daytime symptoms (such as bradykinesia) and nighttime sleep disorders (such as RBD and REM sleep behavior disorder), and cannot effectively monitor cross-cycle symptoms such as "end-of-dose phenomenon" and "morning stiffness".

[0006] The lack of identification of key transition periods: The accurate identification of the "pre-sleep stage" (before falling asleep) and the "first awakening stage" (after waking up and going to bed) is ignored. The former is the key to assessing insomnia (sleep latency), and the latter is the high-incidence period of morning stiffness symptoms. Current technology cannot capture these two stages, resulting in the lack of assessment of related symptoms or reliance on subjective memory.

[0007] The analysis strategy is rigid and the computational efficiency is low: all collected data are processed by a uniform algorithm without taking into account the differences in patients' symptoms under different physiological states (such as deep sleep and awake movement), resulting in wasted computing resources and excessive power consumption, which cannot meet the needs of long-term home real-time monitoring (such as insufficient battery life of wearable devices).

[0008] Imbalanced sleep monitoring solutions: Polysomnography (PSG), the gold standard in laboratories, is bulky and expensive, making it unsuitable for long-term home use; existing home sleep monitoring solutions either only distinguish between "sleep / wake" (coarse staging) or, although attempting fine staging, have complex models (high computational cost) and do not combine staging results with targeted monitoring of sleep disorders (such as PLMS, periodic limb movement disorder).

[0009] The assessment of morning stiffness lacks objective quantitative methods: Clinical assessment of morning stiffness relies entirely on patients' subjective descriptions and scale scores, which are subject to recall bias. Current technology lacks quantitative methods based on objective signals such as electromyography (EMG) and kinetic microscopy (IMU), making it impossible to assess the onset time, duration, and severity of morning stiffness.

[0010] Data is disconnected from clinical decision-making: Existing systems only output raw data or simple event counts (such as the number of falls), and cannot automatically convert them into standardized assessments (such as UPDRS scores), structured reports, or visualized trends. The data has poor interpretability and is difficult to directly assist doctors in diagnosis and treatment adjustments.

[0011] The system lacks personalization and a closed-loop diagnosis and treatment system: it uses fixed algorithms and report templates, which cannot adapt to the asymmetry of patients' symptoms (such as unilateral tremor) or the clinical focus of doctors; at the same time, it has not established a closed loop of "monitoring data - doctor evaluation - medication adjustment", and cannot respond to changes in the condition in a timely manner (such as the decline in drug efficacy). Summary of the Invention

[0012] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0013] A. Integrated Monitoring of Multiple Symptoms Across the Entire Cycle: For the first time, hierarchical state recognition enables continuous 24-hour monitoring of "pre-sleep - nighttime sleep - first awakening - daytime activity," simultaneously covering motor symptoms (frozen gait, bradykinesia) and sleep disorders (RBD, PLMS), revealing cross-cycle symptom correlations (e.g., the correlation between nighttime RBD and daytime bradykinesia). B. Significantly Improved Computational Efficiency and Practicality: Employing a "macro-state recognition - adaptive analysis" strategy, only signals related to the current state are processed (e.g., gait signals are not processed during sleep), reducing computational power consumption by over 40% and extending wearable device battery life to 72 hours, meeting the needs of long-term home monitoring.

[0014] C. Objectification of transitional symptom assessment: Accurately identify the pre-sleep and initial wakefulness stages, providing objective indicators for insomnia (quantification of sleep latency) and morning stiffness (multi-dimensional index), and solving the recall bias problem of traditional subjective assessment.

[0015] D. Sleep monitoring adapted for home scenarios: Based on a simplified sleep staging model of EOG / EMG, it balances accuracy (consistency with PSG staging >85%) and practicality (lightweight device, no professional operation required), while achieving targeted monitoring of sleep disorders.

[0016] E. Deep integration of data and clinical decision-making: Automatically generates standardized UPDRS / PDSS scores and structured reports, supports personalized weight adjustments and medication correlation analysis, and allows physicians to directly adjust treatment plans based on reports, significantly enhancing the clinical value of data.

[0017] F. Improved treatment efficiency through closed-loop diagnosis and treatment: By connecting with HIS through edge-cloud collaboration, the system can achieve "real-time uploading of monitoring data - remote evaluation by doctors - rapid adjustment of treatment plans", reducing the frequency of outpatient visits for patients (from an average of 12 times per year to 6 times) and responding promptly to changes in the condition such as the decline in drug efficacy.

[0018] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages compared with the prior art:

[0019] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the macroscopic state identification of a method and system for comprehensively monitoring daytime and nighttime symptoms of Parkinson's disease according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] It should be noted that the terms "vertical," "horizontal," "up," "down," "left," "right," and similar expressions used in this article are for illustrative purposes only and do not represent the only possible implementation.

[0023] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains; the terminology used herein in the description of the invention is for the purpose of describing particular embodiments only and is not intended to limit the invention; the term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0024] like Figure 1 As shown, one embodiment of this application provides a method for simultaneously monitoring motor symptoms and sleep disorders in Parkinson's disease. The method achieves efficient and accurate state recognition and symptom analysis through hierarchical processing of multimodal sensor data. Specifically, based on real-time clock signals, body position signals, electrooculography (EOG) signals, and mandibular electromyography (EMG) signals, the system first divides the user's macroscopic state, including daytime activity stages, nighttime sleep stages, and transitional stages between the two—pre-sleep stages and initial awakening stages; then, it adaptively invokes corresponding motor and physiological signals according to the user's current state to achieve targeted real-time monitoring, recognition, and analysis of symptoms.

[0025] The system adopts a simplified sleep stage scheme, which mainly distinguishes between wakefulness, REM sleep and non-REM sleep based on electrooculography and electromyography signals. It is suitable for long-term monitoring in a home environment and is easier to use.

[0026] (1) The macroscopic state identification process is as follows:

[0027] During the macroscopic state recognition process, the system continuously monitors body position signals and a real-time clock.

[0028] 1. When it is detected that the user has been in a lying position for a long time (e.g., more than half an hour) during the nighttime period (e.g., 21:00-7:00 the next day), it is preliminarily determined that the user is in a nighttime sleep-related state.

[0029] To further distinguish between the pre-sleep stage and the formal sleep stage, the system performs the following steps:

[0030] 1. Determining the Formal Sleep Start (T_sleep): The system analyzes electrooculography (EOG) and electromyography (EMG) signals in real time. When the system first continuously monitors EOG signals showing a sustained closed-eye state and a significant decrease in slow eye movement frequency, while the root mean square amplitude (RMS) of the EMG signal gradually decreases and stabilizes at sleep levels, the system determines that the user has entered the formal sleep stage and marks this time point as T_sleep.

[0031] 2. Retrospectively locate and verify the pre-sleep stage:

[0032] 2.1 Backtracking Search Start Point: After determining the formal sleep start point (T_sleep), the system initiates a backtracking processing algorithm to analyze data within a preset time window (e.g., 180 minutes) prior to T_sleep. Within this backtracking window, the system searches backwards for the last time point that meets the "fully awake characteristics" as a candidate start point for the presleep stage (T_presleep_candidate). The criteria for determining the "fully awake characteristics" are: clear blinking or eye-opening features in the electrooculography (EOG) signal (e.g., a spike in the EOG signal).

[0033] 2.2 Stage Feature Verification: To confirm that the period between T_presleep_candidate and T_sleep is indeed the presleep stage, the overall characteristics of the signal during this period are further verified. The stability (or peak density) of the oculomotor signal during this period is calculated. If the oculomotor signal during this period mainly exhibits stable characteristics under closed-eye conditions (such as low-frequency, low-amplitude fluctuations), and only intermittently interspersed with peak signals related to blinking or opening eyes, then the verification is successful, and T_presleep_candidate is officially determined as the starting point of the presleep stage (T_presleep).

[0034] 2.3 Phase Confirmation: The time interval between the verified start point T_presleep and the end point T_sleep (ΔT_sleep_latency = T_sleep - T_presleep) is determined as the duration of the presleep phase.

[0035] 2. If the system detects that the body position changes from lying down to sitting or standing, and does not return to lying down after a long period of time (e.g., 40 minutes), and the real-time clock is in the preset early morning period (e.g., 05:00 to 10:00), the system determines the event as a "morning event" and marks this time point (T_stand) as the official start of the daytime activity phase.

[0036] To further distinguish between the initial awakening stage and the daytime activity stage, the system performs the following steps:

[0037] 1. Retrospectively locate the candidate start time point of the initial awakening stage (T_wake_candidate): Activate the retrospective processing algorithm to automatically analyze physiological signal data, including electrooculography (EOG) and electromyography (EMG) signals, within a preset time window (e.g., 60 minutes) prior to T_stand. Within this retrospective window, search backwards for the most recent time point that meets the "sudden awakening characteristics" as the candidate start point of the initial awakening stage (T_wake_candidate). The criteria for determining the "sudden awakening characteristics" are: a sustained instantaneous increase in blink frequency in the EOG signal, and / or a step increase in the root mean square amplitude of the mandibular and limb EMG signals.

[0038] 2. Stage Feature Verification: To confirm that the time period between T_wake_candidate and T_stand is indeed the initial wake-up stage, the overall characteristics of the signal during this period are further verified. The overall trend and characteristics of the multimodal signal within this candidate time period are calculated and analyzed. Verification can only pass if the following conditions are met simultaneously:

[0039] During this period, the user should remain in a supine position until the T_stand time point, at which point they should transition to a sitting / standing position. This indicates that the user is awake but has not yet begun daytime activities. Simultaneously, the electrooculography (EOG) signal should exhibit the pattern of a normal awake state, which may show steady, low-frequency fluctuations with eyes closed, or intermittent characteristic spikes related to eye opening and blinking. The root mean square (RMS) amplitude of the chin-EMG signal should remain consistently above the awake baseline level during sleep. Furthermore, the EOG and EMG signal data should not reproduce characteristics predicting sleep onset; that is, the EOG signal should not show continuous slow eye movements, and the EMG signal tone should not show a progressive decrease and stabilize at sleep levels.

[0040] The above verification criteria aim to exclude the possibility of the user falling asleep again during this period, such as a sudden awakening followed by re-falling asleep, ensuring that the identified period is a continuous "initial awakening" state. If the above feature verification passes, then T_wake_candidate is officially determined as the starting point (T_wake) of the initial awakening stage.

[0041] 3. Phase Confirmation: The time interval (ΔT_stand_latency = T_stand - T_wake) between the verified start point T_wake and the end point T_stand is determined as the duration of the initial awakening phase.

[0042] (2) Adaptive symptom analysis based on macro-state labels

[0043] Based on successfully identifying and classifying the user into any macroscopic state—daytime activity, nighttime sleep, pre-sleep, or initial awakening—the system adaptively activates the most relevant specific symptom analysis submodule according to the state label. This strategy significantly optimizes computational resource allocation, improves system processing efficiency and real-time performance, and reduces overall power consumption by focusing on the symptom set with the highest probability in the current state and selectively processing the related sensor data streams.

[0044] 2.1 Analytical strategies for symptom identification during daytime activity phases:

[0045] Real-time activity status classification: Based on body position signal data, limb acceleration and angular velocity signal data, a lightweight machine learning model (such as support vector machine or decision tree) is used to classify the user's current activity status in real time. The main categories include: walking, standing, sitting and lying down.

[0046] Symptom monitoring while walking and standing:

[0047] ① In this state, lower limb acceleration signal data, angular velocity signal data, and surface electromyography signal data are prioritized for analysis. Time-domain (e.g., signal amplitude range, variance) and frequency-domain feature extraction methods are used to identify frozen gait and panicked gait. To further refine the diagnosis, body position change data (e.g., turning angular velocity) are combined to differentiate between starting-type frozen gait and turning-type frozen gait.

[0048] ② Based on the burst patterns of lower limb electromyographic signals, the impact characteristics of acceleration signals, and abnormal changes in angular velocity signals, the system detects fall events in real time. Furthermore, by analyzing the temporal correlation between freezing events and fall occurrences, as well as electromyographic patterns, the system performs attribution analysis on falls, distinguishing between freezing-related falls and non-freezing-related falls (such as those caused by postural imbalance).

[0049] ③ Synchronously analyze IMU signals from the limbs to extract features such as range of motion and average speed to quantify bradykinesia. Simultaneously, analyze data on small-amplitude swaying (forward leaning, backward leaning, left and right swaying) related to the center of posture and the amplitude-frequency characteristics of limb swaying to identify postural instability.

[0050] Symptom monitoring in sitting and lying positions:

[0051] In this state, electrooculography (EOG) and mandibular muscle (EMG) data are prioritized for analysis. Daytime sleep events are identified by detecting a persistently abnormally high proportion of slow eye movements in the EOG data, combined with a significant decrease in mandibular tension (i.e., the root mean square amplitude of the EMG signal is below the wakefulness baseline threshold). Each identified sleep event is precisely labeled with its start time, end time, and duration, and a sleep log is generated. To avoid misdiagnosing normal afternoon naps as pathological hypersomnia, the final clinical assessment must be made by a physician based on a comprehensive judgment considering the event's timing, frequency, and the patient's subjective report.

[0052] 2.2 Analytical strategies for identifying symptoms during nighttime sleep stages:

[0053] Sleep staging: First, based on electrooculography (EOG) data and mandibular myocardial electromyography (EMG) data, a lightweight machine learning model (such as a hidden Markov model or random forest) is used to automatically staging sleep, identifying the wakefulness stage, non-rapid eye movement (NREM) sleep stage, and rapid eye movement (REM) sleep stage.

[0054] Stage-specific sleep disorder monitoring:

[0055] During this REM sleep phase, the system automatically activates the analysis of limb acceleration signal data, angular velocity signal data, electromyography (EMG) signal data, and postural signal data to monitor abnormal movements of the limbs and trunk. Typical symptoms of REM sleep behavioral disorders are identified by detecting brief bursts of limb EMG signal amplitude exceeding a preset threshold (e.g., more than twice the background EMG activity level), and / or corresponding limb twitches in the limb acceleration and angular velocity signals.

[0056] Pre-sleep period and non-rapid eye movement sleep period:

[0057] The study focuses on monitoring lower limb acceleration, angular velocity, and electromyography signals to identify frequent and irregular twitching movements in the legs, thus aiding in the diagnosis of subjective discomfort-related movements in restless legs syndrome.

[0058] ③ Non-rapid eye movement (NREM) sleep: The system also focuses on monitoring lower limb acceleration, angular velocity, and electromyography (EMG) signals, but uses different algorithms to detect periodic limb movements. For example, the system identifies a series of leg movements with consistent duration and intervals to aid in the diagnosis of periodic leg movements.

[0059] 2.3 Overall Sleep Quality Monitoring and Analysis Strategy:

[0060] ① Sleep latency: The time interval between the start of the presleep stage T_presleep and the start of the formal sleep stage T_sleep is marked as the sleep latency. The duration is calculated by the formula ΔT_sleep_latency=T_sleep-T_presleep. This duration can be directly used to objectively quantify the degree of difficulty a user has in falling asleep, i.e. the severity of insomnia symptoms.

[0061] ②Nighttime awakening: Nighttime awakening is detected by monitoring sudden changes in electrooculography (EOG) and mandibular electromyography (EMG) data (such as a sudden increase in blink frequency and a step increase in mandibular EMG RMS value).

[0062] ③ Nighttime urination events: Identify and record nighttime urination events by monitoring the behavior of changing from a supine position to a standing / walking position and returning to a supine position within a certain time window (e.g., 30 minutes).

[0063] Turning-over events: By analyzing data from body position change sensors, nighttime turning-over events are identified as an auxiliary assessment indicator of sleep fragmentation.

[0064] 2.4 Strategies for Identifying and Analyzing Morning Stiffness in Parkinson's Disease

[0065] Morning stiffness, also known as dystonia, is a common "wearing-off phenomenon" in the progression of Parkinson's disease. It is characterized by generalized rigidity and difficulty in movement for a period of time after waking up in the morning. Some patients experience spontaneous relief within minutes to half an hour, while others require medication. Its pathophysiological basis is closely related to insufficient dopaminergic nerve conduction leading to dystonia and insufficient drug storage in the central nervous system due to the long nighttime hours. This symptom often recurs after nighttime sleep, and is most pronounced during the transition from sleep to daytime activity (what this system identifies as the "initial awakening stage"). Traditional clinical assessment relies primarily on patient subjective descriptions and scale scores, which are limited by recall bias and quantification inaccuracies.

[0066] This application leverages its multimodal sensing and state recognition capabilities to provide an innovative solution for achieving objective, non-intrusive, and quantitative monitoring of morning stiffness symptoms. Specifically, after successfully identifying the time range [T_wake, T_stand] of the initial awakening stage, the morning stiffness quantitative analysis submodule is automatically activated and executes the following analysis process:

[0067] Preliminary calculation of the morning stiffness time window: The system first calculates the time interval (ΔT = T_stand - T_wake) between the physiological wakefulness time (T_wake) and the behavioral wakefulness time (T_stand). This interval ΔT serves as a preliminary suspected indicator of the duration of morning stiffness, and an abnormally prolonged interval (such as exceeding the baseline of healthy peers) suggests the possible presence of morning stiffness symptoms.

[0068] Multimodal signal verification and severity quantification: To confirm that the motor impairment within the time window ΔT is indeed "morning stiffness" (i.e., muscle rigidity) and may be accompanied by bradykinesia, and to rule out cases where it is merely due to resting in bed after waking up, the system further analyzes the limb acceleration, angular velocity, and electromyographic signals during this period (T_wake to T_stand), extracting the following features for cross-validation and quantification:

[0069] a. Verification and quantification of myotonia based on electromyographic signals: The system calculates the average amplitude or integrated electromyographic value of the limbs during the specified time period. A significantly elevated and sustained level of electromyographic activity compared to the user's normal daytime waking state provides objective electrophysiological evidence of pathological myotonia. The amplitude or energy of this electromyographic activity can serve as a direct indicator for quantifying the severity of myotonia.

[0070] b. Verification and Quantification of Bradykinesia Based on Motion Signal Data: Simultaneous analysis of triaxial acceleration and angular velocity signals of the limbs during this time period. Motor function is objectively assessed by calculating temporal and frequency domain characteristics such as limb amplitude range, average velocity, motion entropy, or zero-crossing rate. Significantly lower than normal amplitude and velocity indicate the presence of bradykinesia. The degree of reduction in amplitude and slowness of movement can serve as auxiliary indicators for quantifying the severity of bradykinesia.

[0071] c. Fusion Assessment and Index Generation: Finally, the system integrates multi-dimensional information such as time indicators (ΔT), electromyographic characteristics (MAV / iEMG), and kinematic characteristics (ROM, velocity), and generates a fused morning stiffness severity index through a pre-set weighted algorithm or machine learning model (such as logistic regression). This index not only achieves a binary judgment of "presence / absence" of morning stiffness symptoms, but also realizes continuous, objective, and multi-dimensional quantification of its severity, providing high-value data for clinical efficacy assessment and disease progression monitoring.

[0072] All the quantitative indicators generated by the above analysis (ΔT, amplitude of increased electromyographic activity, percentage of decreased range of motion, and fusion severity index) were written into a structured assessment report with corresponding time period markers for doctors to use as a reference for final clinical diagnosis and decision-making.

[0073] Through the above-mentioned hierarchical processing and state-dependent analysis strategy, this invention can significantly reduce system computation and communication overhead while ensuring the accuracy of symptom identification. It is suitable for long-term, wearable multi-symptom monitoring scenarios of Parkinson's disease, and provides objective and continuous data support for disease assessment and treatment adjustment.

[0074] On the other hand, based on the above-mentioned method for simultaneously monitoring motor symptoms and sleep disorders of Parkinson's disease, this application proposes a system that provides objective structured data to assist in the diagnosis of the prevalence and progression of Parkinson's disease.

[0075] Motor symptoms are rated based on the monitoring and identification of Parkinson's disease motor symptoms (location, on / off time, and duration of one or more Parkinson's disease motor symptoms), and Parkinson's sleep disorders are scored based on Parkinson's disease sleep disorder information (various types of sleep disorders in Parkinson's patients, their occurrence time, and duration), and a structured report is generated for doctors to review to assist in diagnosis.

[0076] The system specifically includes the following functional modules and implementation process:

[0077] 1. Symptom Severity Rating and Scoring

[0078] The system performs quantitative rating and scoring based on the monitored and identified information on Parkinson's disease motor symptoms and sleep disorders.

[0079] Motor symptom rating: For one or more identified Parkinson's disease motor symptoms (such as tremor, rigidity, bradykinesia, postural instability, frozen gait, REM sleep disorder, restless legs syndrome, etc.), severity can be rated according to internationally accepted standards (such as the Unified Parkinson's Disease Rating Scale, UPDRS) based on quantitative characteristics such as location, frequency, onset and end time, duration, amplitude, and frequency. Rating can employ machine learning regression models (such as support vector regression, random forest regression, XGBoost regression, etc.) to map extracted multidimensional features to continuous severity scores (e.g., 0-4 points) to achieve an objective and quantitative assessment of symptom severity.

[0080] Sleep Disorder Scoring: For identified sleep disorders (such as insomnia, REM sleep behavior disorder, restless legs syndrome / periodic limb movement disorder, excessive daytime sleepiness, etc.), scores are assigned based on indicators such as type, time of occurrence, frequency, and duration. For example, based on parameters such as total sleep duration, sleep latency, number of awakenings, duration and proportion of each sleep stage, and indices of specific sleep events (such as REM sleep behavior disorder movements, restless legs syndrome), machine learning models (such as random forest regression) can predict scores corresponding to standard sleep scales (such as the PDSS).

[0081] 2. Comprehensive assessment and structured report generation

[0082] The system generates structured reports for clinicians by comprehensively analyzing and visualizing the rating and scoring results.

[0083] a. Customizable comprehensive scoring and weighting adjustment mechanism

[0084] To enable more personalized assessments, the system supports a customizable comprehensive scoring system.

[0085] The comprehensive score is generated by integrating the rating data of various motor symptoms and their frequency of occurrence and total duration per day, activity patterns (daytime sleepiness, etc.), and nighttime sleep status. Each dimension has a certain weight, and the comprehensive score is generated by weighted integration. Within each dimension, the weight of each item is also different.

[0086] For example, in the motor symptoms dimension, symptoms with higher weights include those on the affected side, those with large recent fluctuations and high frequency, and those with high severity ratings. The weights of each item and dimension are determined through development experiments, clinical experience, and expert consensus.

[0087] For example, it supports the selection of monitoring indicators, that is, doctors or researchers can independently select the core monitoring indicator set that needs to be focused on from all the symptoms and parameters identified by the system based on clinical concerns or research purposes (such as focusing on tremors and rigidity of the left limbs, or focusing on REM sleep disorder and daytime sleepiness).

[0088] For example, the system supports dynamic weight adjustment, allowing for dynamic weight adjustments based on clinical rules when generating a comprehensive disease score. Given that Parkinson's disease often has a unilateral onset and asymmetrical symptoms, the algorithm automatically assigns higher weights to body parts on the side with more severe symptoms. More importantly, the system provides a physician annotation interface, allowing doctors to manually adjust the weights of different symptoms in the comprehensive score based on clinical judgment (e.g., assigning higher weights to frequently occurring "frozen gait" to highlight its importance to the patient's fall risk), thereby making the generated comprehensive score more closely aligned with the individual patient's situation and current core clinical issues.

[0089] b. Data visualization generation:

[0090] Based on the above identification and quantification results, the system generates quantitative scoring curves (such as "tremor severity-time curve") reflecting the change of symptom severity over time in different locations, according to symptom type (such as tremor, rigidity, bradykinesia, dyskinesia, REM sleep behavior disorder, restless legs syndrome, excessive daytime sleepiness, insomnia, nighttime awakening, getting up and walking, etc.) and / or location of occurrence. In addition, it can further generate proportion charts or pie charts reflecting the distribution of the patient's daytime behavior (such as the proportion of time spent sitting, standing, walking, jogging, sleeping, etc.).

[0091] c. Structured report statistics and generation:

[0092] The system automatically generates structured reports, including but not limited to: key indicators for each symptom or event, such as location, start and end time, frequency, average duration, and maximum / average severity; key quantitative indicators, such as average gait and typical tremor frequency; daytime and nighttime trends of key symptoms (e.g., morning stiffness, the phenomenon where muscle rigidity is most severe upon waking in the morning); historical data comparison and analysis, such as comparing current monitoring data with historical monitoring data to calculate the percentage change in symptom frequency or severity and highlighting significant changes in the report; preliminary analysis of symptom changes based on medication time-related factors, such as marking medication time points through patient interaction and marking medication time points on the symptom fluctuation curve to observe symptom fluctuations with medication; analysis of reduced daytime activity; analysis of daytime sleepiness duration and frequency; and interactive review options for clinicians (e.g., zooming in and out of the timeline, focusing on specific symptom events).

[0093] d. Report Output

[0094] The generated clinical reports, which include visualizations and structured data, can be output to a designated storage location or transmitted to authorized users (such as clinician workstations or hospital information systems) via a secure interface to display the relevant content and compare it with previous or next monitoring, or to perform further processing.

[0095] As one example, in the Parkinson's diagnosis process, the acquired indicator data and the results of Parkinson's symptom monitoring and processing can be combined to process the corresponding assessment report. The quantitative curves, symptom statistics and structured reports generated by the system provide doctors with objective and quantitative assessment basis for Parkinson's motor symptoms, assisting in diagnostic decision-making and the formulation of initial treatment and medication plans.

[0096] As another exemplary embodiment, for patients diagnosed with Parkinson's disease, the system is worn regularly (e.g., monthly or as prescribed by a doctor) for monitoring. The system integrates new data with historical databases, automatically generating disease progression reports that highlight symptom trends (e.g., worsening or improvement). This report is transmitted to doctors via a secure telemedicine platform (such as an internet hospital system) for review, archiving, and as a supplementary basis for diagnostic decisions and treatment planning. Doctors can then perform the following actions:

[0097] Review the preliminary report and quantitative curves generated by the system;

[0098] As needed, review the initial signal of each physical channel (e.g., if you suspect an electrode contact problem or are verifying a specific event);

[0099] The disease progression and drug efficacy were comprehensively assessed by combining the patient's subjective feelings, clinical observations, and reported data.

[0100] Adjust the medication regimen accordingly (such as the type of medication, dosage, and timing of administration).

[0101] e. System optimization and personalized configuration

[0102] Continuous Model Optimization: The system uses its powerful computing capabilities to continuously train and optimize the machine learning or deep learning models used for symptom recognition and scoring by anonymizing and aggregating desensitized multi-user data (e.g., hyperparameter tuning, feature selection). The optimized models can be periodically distributed to user terminals, thereby enabling the overall system's analytical performance to iteratively improve over time.

[0103] Personalized configuration management: The system supports the creation of personalized monitoring and analysis plans for different patients. These plans include selected monitoring indicators, weight preference settings, report templates, etc., and can be saved and applied with a single click, achieving precise monitoring and management tailored to individual patients.

[0104] In summary, this system constructs a clinical intervention-driven management loop of "monitoring-analysis-reporting-physician evaluation-treatment adjustment," aiming to overcome the limitations of the short-term and subjective nature of traditional clinical assessments. It provides objective and comprehensive data indicators to support the diagnosis of Parkinson's disease. At the same time, it helps patients diagnosed with Parkinson's disease to achieve more timely medical intervention when drug efficacy declines or the disease progresses rapidly, thereby improving treatment outcomes, patients' quality of life, and medical efficiency.

[0105] When in use, this device has the following main advantages:

[0106] 1. A hierarchical macroscopic state recognition method based on multimodal signals and real-time clocks

[0107] By continuously monitoring body position signals and a real-time clock, the system automatically divides the user's state into daytime activity stage, nighttime sleep stage, pre-sleep stage, and initial awakening stage.

[0108] The formal sleep onset (T_sleep) is determined by using the characteristics of electrooculography (EOG) and electromyography (EMG) signals of the mandible (e.g., closed eyes and decreased EMG amplitude).

[0109] A backtracking algorithm is used to search backward within a preset time window and verify the start point of the presleep stage (T_presleep) and the start point of the initial wake-up stage (T_wake) to ensure the accuracy of the stage division.

[0110] 2. State-adaptive symptom identification and analysis strategies

[0111] The system automatically activates the corresponding symptom analysis submodule based on the currently identified macroscopic state (such as daytime activity, nighttime sleep, etc.), and only processes the sensor data streams most relevant to that state, significantly reducing computing resource consumption and power consumption.

[0112] During daytime activity: Prioritize analysis of limb IMU and surface electromyography signals to identify frozen gait, panicked gait, fall events, bradykinesia, and postural instability.

[0113] During nighttime sleep: Based on simplified sleep stages (wakefulness, REM, NREM), limb movements are monitored during REM sleep to identify RBD (rapid eye movement sleep behavior disorder), and periodic limb movements (PLMS) and restless legs syndrome (RLS) related activities are monitored during NREM sleep.

[0114] 3. Objective quantitative analysis methods for morning stiffness (morning dystonia)

[0115] During the initial awakening phase (T_wake to T_stand), the presence and severity of morning stiffness were verified through cross-validation using multimodal signals (electromyography, acceleration, angular velocity):

[0116] Electromyography (EMG) signals were used to detect muscle rigidity (iEMG values ​​were significantly higher than baseline);

[0117] Motion signals are used to quantify motion sluggishness (reduction in motion amplitude and speed);

[0118] By integrating time window (ΔT), electromyographic features, and kinematic features, a morning stiffness severity index is generated, achieving multi-dimensional objective quantification.

[0119] 4. Automatic monitoring and index extraction of sleep quality and sleep disorders

[0120] Automatically calculates sleep latency (T_presleep to T_sleep), identifies nighttime awakenings (based on EOG / EMG mutations), nighttime waking events (body position changes + time window judgment), and turning events (body position sensor) as objective indicators of sleep fragmentation and insomnia.

[0121] During sleep, REM and NREM phases are distinguished, and different signal processing strategies are applied to RBD, PLMS, and RLS respectively.

[0122] 5. Automated rating and scoring mechanism for symptom severity

[0123] Machine learning models (such as SVR and random forest regression) are used to map multimodal features to continuous scores on international standard scales (such as UPDRS and PDSS) to achieve objective quantitative rating of motor symptoms and sleep disorders.

[0124] Supports dynamic weight adjustment: Based on symptom location, frequency, severity, and doctor's manual annotations, the weight of different symptoms in the comprehensive score is adaptively adjusted to improve personalized assessment capabilities.

[0125] 6. Structured Report Generation and Visualization System

[0126] Automatically generate visualized clinical reports that include time-series curves (such as tremor severity-time curves), behavioral distribution scales, and event statistics (frequency, duration, location).

[0127] It supports historical data comparison and medication time correlation analysis, and can output to doctor workstations or hospital information systems through a secure interface.

[0128] 7. Closed-loop personalized management and system optimization mechanism

[0129] It supports configuring personalized monitoring plans (indicator selection, weight setting, report templates) for different patients.

[0130] The system continuously optimizes machine learning models by anonymously aggregating multi-user data and supports remote model updates, thereby constantly improving the accuracy of recognition and scoring.

[0131] Construct a closed-loop clinical intervention system encompassing "monitoring, analysis, reporting, physician evaluation, and treatment adjustment" to improve the efficiency of long-term disease management.

[0132] 8. Multimodal signal fusion and lightweight algorithm design

[0133] In resource-constrained wearable devices, lightweight machine learning models (such as decision trees, HMMs, and SVMs) are used to achieve real-time classification and recognition.

[0134] By extracting signal features at different levels (time domain, frequency domain, and entropy features) and fusing multi-sensor data (IMU+EMG+EOG+body position), the computational complexity is controlled while ensuring accuracy.

[0135] Taking a 65-year-old Parkinson's disease patient (diagnosed 3 years ago, with morning stiffness, freezing gait, and RBD symptoms) as an example, the 72-hour home monitoring process is as follows:

[0136] Device Wearing and Signal Acquisition: Before going to sleep, the patient wears the wrist device and EOG patch, and the mattress sensor is placed in a designated position on the mattress; the APP automatically synchronizes the devices and collects EOG (256Hz), EMG (500Hz), IMU (50Hz), body position (1Hz) and clock signals. The raw signals are filtered by the APP (50Hz notch filter, 0.1-50Hz bandpass filter) and characteristic data (such as blink frequency of EOG and average acceleration value of IMU) are transmitted to the cloud every 5 minutes.

[0137] Macroscopic state identification:

[0138] At night (21:30): The mattress sensor detected a continuous 30-minute lying position, and the clock was in the nighttime period, which initially determined the nighttime sleep-related status; at 22:15, EOG showed a stable characteristic with eyes closed (slow eye movement frequency <5 times / minute), and the mandibular EMG-RMS dropped to 50% of the wakefulness baseline, marking T_sleep = 22:15; reviewing 180 minutes of data, it was found that EOG showed the last blink peak at 21:45, verifying that EOG was mainly stable with eyes closed during this period, and T_presleep = 21:45 was determined, with the pre-sleep stage lasting 30 minutes (ΔT_sleep_latency = 30 minutes, indicating mild insomnia).

[0139] Early morning (6:00): The mattress sensor detected a shift from a supine to a sitting position, which lasted for 40 minutes without returning to a supine position. The clock was in the early morning period, and T_stand = 6:40 was marked. Reviewing 60 minutes of data, it was found that at 6:10, the EOG blink rate increased from 5 times / minute to 20 times / minute, and the EMG-RMS stepped up to the wakefulness baseline. This confirmed that the body position remained supine during this period, and T_wake = 6:10 was determined. The initial awakening stage lasted for 30 minutes (ΔT = 30 minutes).

[0140] Adaptive symptom analysis:

[0141] During the daytime activity phase (6:40-21:30): The APP loads a decision tree model to classify activity states (walking 25%, standing 15%, sitting 45%, lying down 15%). At 10:15, when the patient was walking, the IMU detected a sudden drop in acceleration (<0.1g) and high-frequency tremor (8-12Hz), and EMG showed a continuous burst, which was identified as "starting-type frozen gait" (lasting 5 seconds). At 14:30, when the patient was sitting, the EOG slow eye movement accounted for 70%, and the EMG-RMS was lower than the baseline, which was identified as "daytime sleep event" (lasting 20 minutes).

[0142] Nighttime sleep stage (22:15-6:10): A random forest model is loaded in the cloud to divide sleep into REM sleep (2 hours), NREM sleep (5 hours), and wakefulness sleep (1 hour); during the REM sleep at 2:30 am, the IMU of the four limbs detected 3 limb twitches, and the EMG amplitude was 3 times that of the background, which was identified as "RBD symptoms"; during the NREM sleep at 4:00 am, the IMU of the lower limbs detected 8 regular leg movements per hour, which was identified as "PLMS".

[0143] Initial awakening stage (6:10-6:40): The logistic regression model is loaded in the cloud and ΔT = 30 minutes is calculated. The MAV of EMG is 40% higher than the baseline of wakefulness (indicating muscle rigidity), and the ROM of IMU is 30% lower than the normal level (indicating bradykinesia). The morning stiffness severity index is generated = 7 points (moderate morning stiffness).

[0144] Quantitative assessment and report generation:

[0145] Quantitative scoring: Motor symptoms were rated according to UPDRS (frozen gait 3 points, bradykinesia 2 points), and sleep disorders were rated according to PDSS (RBD 5 points, PLMS 6 points); the doctor adjusted the weight of frozen gait through the web (from 20% to 40%), and the personalized comprehensive score was 6.5 points.

[0146] Structured Report: The APP generates a “Morning Stiffness Index - 72-Hour Curve” (index score ≥ 6 points for the period from 6:10 to 6:40), a “Daytime Activity Ratio Pie Chart” (seated position has the highest proportion), and a “Medication Correlation Chart” (the incidence of frozen gait decreased by 50% 2 hours after medication); the report is automatically uploaded to the hospital’s HIS system.

[0147] Closed-loop diagnosis and treatment: After reviewing the report, the doctor determined that the patient's morning stiffness and nocturnal RBD symptoms had worsened, and adjusted the medication regimen (increasing the levodopa dose and advancing the administration time by 30 minutes); the adjusted regimen was pushed to the patient via the APP, and the next round of monitoring (1 month later) showed that the morning stiffness index dropped to 4 points and the incidence of RBD decreased by 60%.

[0148] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Anyone skilled in the art can make various modifications and alterations without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be determined by the claims.

Claims

1. A method for comprehensively monitoring daytime and nighttime symptoms of Parkinson's disease, characterized in that, Includes the following steps: 1) Multimodal signal acquisition: Synchronously acquire the user's real-time clock signal, body position signal, electrooculography (EOG) signal, electromyography (EMG) signal of the mandible and limbs, and inertial measurement unit (IMU) signal, wherein the IMU signal includes acceleration signal and angular velocity signal; 2) Hierarchical Macroscopic State Recognition: Based on the real-time clock signal, body position signal, EOG signal, and EMG signal collected in step 1), the user's physiological state is divided into the daytime activity stage, nighttime sleep stage, pre-sleep stage, and initial awakening stage. The specific division process is as follows: a) When the body is detected to be in a supine position for a preset duration (≥30 minutes) and the real-time clock is in the nighttime period (e.g., 21:00-7:00 the next day), it is initially determined to be a nighttime sleep-related state. By analyzing the stability of the EOG signal with eyes closed (low-frequency, low-amplitude fluctuations) and the gradual decreasing trend of the root mean square amplitude (RMS) of the EMG signal, the formal sleep start point (T_sleep) is determined. The EOG / EMG signals within the preset time window (e.g., 180 minutes) before T_sleep are reviewed, and the last time point that meets the "fully awake characteristics" (EOG blinking spike) is searched as the candidate start point of the presleep stage (T_presleep_candidate). After verifying that the EOG in the candidate period is mainly characterized by stability with eyes closed, the start point of the presleep stage (T_presleep) and the presleep stage are determined. b) When the body position changes from supine to sitting / standing and does not return to supine for a preset duration (≥40 minutes), and the real-time clock is in the early morning period (e.g., 05:00-10:00), mark the start point of the daytime activity stage (T_stand); backtrack the EOG / EMG signals within the preset time window (e.g., 60 minutes) before T_stand, and search for time points that meet the "sudden awakening characteristics" (instantaneous increase in blink frequency or step increase in EMG-RMS) as candidate start points of the initial awakening stage (T_wake_candidate). After verifying that the body position remains supine and the EOG / EMG is in a wakeful mode during the candidate time period, determine the start point of the initial awakening stage (T_wake) and the initial awakening stage. 3) Adaptive Symptom Analysis Based on Macro-State: Based on the macro-states defined in step 2), the corresponding symptom analysis sub-module is adaptively activated, processing only signals relevant to the current state to identify Parkinson's disease symptoms. Specifically, this includes: a) Daytime Activity Phase: Activate the daytime symptom analysis submodule. Based on IMU and EMG signals, use a lightweight machine learning model (support vector machine or decision tree) to classify activity states (walking, standing, sitting, lying down). For walking / standing states, analyze the time domain (amplitude range, variance) and frequency domain features of the signals to identify frozen gait (distinguishing between starting and turning gait) and hurried gait. Detect and attribute fall events (frozen correlation / non-frozen correlation) through electromyographic burst patterns and abnormal motion signals. For sitting / lying states, analyze the proportion of slow eye movements (EOG) and EMG-RMS threshold to identify daytime sleep events. b) Nighttime Sleep Stage: Activate the nighttime symptom analysis submodule and use a lightweight machine learning model (Hidden Markov Model or Random Forest) based on EOG / EMG signals to complete sleep stages (wakefulness, rapid eye movement (REM) sleep, and non-rapid eye movement (NREM) sleep). For REM sleep, monitor limb IMU / EMG signals to identify rapid eye movement sleep behavior disorder (RBD), and for NREM sleep, monitor lower limb IMU / EMG signals to identify periodic limb movements (PLMS) and restless legs syndrome (RLS) related activities. At the same time, calculate sleep latency (ΔT_sleep_latency = T_sleep - T_presleep), and identify nighttime awakenings, nighttime urination events, and turning-over events. c) Initial Awakening Stage: Activate the morning stiffness analysis submodule, calculate the time interval (ΔT) from T_wake to T_stand; analyze the mean amplitude value (MAV) / integrated electromyography value (iEMG) of EMG signals during this period to verify muscle rigidity, and analyze the range of motion (ROM) / mean velocity of IMU signals to verify bradykinesia; fuse ΔT, MAV / iEMG, and ROM / velocity to generate a morning stiffness severity index; 4) Symptom Quantitative Assessment: Based on the symptom identification results in step 3), a machine learning model (support vector regression, random forest regression, or XGBoost regression) is used to map multimodal features to international standard scale scores. Motor symptoms are rated for severity using the Unified Parkinson's Rating Scale (UPDRS), and sleep disorders are scored using the Parkinson's Disease Sleep Scale (PDSS). 5) Structured report generation: Integrate the quantitative assessment results from step 4) to generate a structured report that includes symptom severity-time curves, daytime activity status ratio charts, statistics on symptom occurrence location / frequency / duration, historical data comparisons, and medication time correlation analysis.

2. The method according to claim 1, characterized in that, In step 3)a), the fall event attribution analysis specifically involves: distinguishing between freeze-related falls and non-freeze-related falls (such as those caused by postural imbalance) by comparing whether there are frozen gait characteristic signals (such as high-frequency tremors in IMU signals and sudden drops in motion amplitude) and abnormal rigidity patterns in EMG signals within 10-30 seconds before the fall occurs.

3. The method according to claim 1, characterized in that, In step 3)b), the criteria for identifying nighttime wake-up events are: monitoring a change in body position from lying down to standing / walking, and returning to lying down within a preset time window (e.g., 30 minutes); the turning-over event is determined by the frequency of body position changes and angle threshold (e.g., ≥30°) collected by the body position sensor.

4. The method according to claim 1, characterized in that, In step 3)c), the criteria for verifying muscle rigidity are: the MAV / iEMG value of the EMG signal is ≥30% higher than the user's baseline value in a normal waking state during the day; the criteria for verifying bradykinesia are: the ROM / average velocity of the IMU signal is ≥25% lower than the normal level.

5. The method according to claim 1, characterized in that, It also includes a personalized weighting adjustment step: allowing doctors to manually adjust the weight of different symptoms in the quantitative assessment based on the asymmetry of the patient's symptoms (such as more pronounced symptoms on one side of the body) and clinical concerns (such as focusing on monitoring frozen gait) to optimize the overall score results.

6. The method according to claim 1, characterized in that, It also includes continuous model optimization steps: anonymizing and aggregating multi-user desensitized monitoring data, performing hyperparameter tuning and feature selection on the machine learning model used for symptom identification and quantitative assessment, and regularly updating the model to improve identification accuracy and generalization ability.

7. A system for comprehensively monitoring daytime and nighttime symptoms of Parkinson's disease, for implementing the method according to any one of claims 1-6, characterized in that, include: Signal acquisition module: Composed of a real-time clock unit, body position sensor, EOG sensor, jaw and limb EMG sensor, and IMU sensor (including accelerometer and gyroscope), used to synchronously acquire multimodal signals; Macroscopic state recognition module: electrically connected to the signal acquisition module, including a pre-sleep recognition unit and a first awakening recognition unit; the pre-sleep recognition unit is used to determine T_sleep and T_presleep and divide the pre-sleep stage, and the first awakening recognition unit is used to determine T_wake and T_stand and divide the first awakening stage, thus realizing the division of daytime activity / nighttime sleep / pre-sleep / first awakening stages as a whole; Adaptive symptom analysis module: electrically connected to the macroscopic state recognition module, including a daytime symptom analysis submodule, a nighttime symptom analysis submodule, and a morning stiffness analysis submodule; the daytime symptom analysis submodule performs symptom recognition and fall attribution during daytime activity phases, the nighttime symptom analysis submodule performs sleep staging and sleep disorder recognition, and the morning stiffness analysis submodule performs quantitative analysis of morning stiffness; Quantitative assessment module: electrically connected to the adaptive symptom analysis module, including a model inference unit and a weight adjustment unit; the model inference unit performs the mapping rating / scoring of symptoms and standard scales, and the weight adjustment unit receives personalized weight parameters input by the doctor and adjusts the quantitative rules; Report generation module: electrically connected to the quantitative assessment module, including a visualization unit and a historical comparison unit; the visualization unit generates symptom curves and percentage charts, the historical comparison unit calculates the percentage change in symptoms between current and historical data, and finally outputs a structured report; Edge-cloud collaborative processing module: The edge device (wearable device) performs signal filtering, preliminary feature extraction and emergency event (such as fall) detection, while the cloud server performs complex model inference, data storage and model optimization and update.

8. The system according to claim 7, characterized in that, It also includes a real-time reminder module: electrically connected to the adaptive symptom analysis module, when a fall event or a morning stiffness severity index ≥8 points (out of 10 points) is detected, an alarm is issued through the vibrator and speaker of the wearable device, and a reminder message containing the event time, location and symptom data is sent to the preset caregiver / doctor terminal.

9. The system according to claim 7, characterized in that, The sensors of the signal acquisition module are integrated into the wearable device, including an IMU / EMG sensor worn on the wrist, an EOG sensor worn on the eyes, and a body position sensor placed on the mattress, to achieve non-intrusive signal acquisition.

10. The system according to claim 7, characterized in that, The report generation module also supports secure data interaction with the Hospital Information System (HIS) / Electronic Medical Record System (EMR), and can automatically upload structured reports to the doctor's workstation for doctors to use for diagnostic decision-making and treatment plan adjustment, forming a clinical closed loop of "monitoring-analysis-reporting-doctor evaluation-treatment adjustment".

Images