An intelligent method and system for assessing functional impairment in Parkinson's disease.

Abstract

This invention discloses an intelligent method and system for assessing functional impairment in Parkinson's disease, belonging to the field of assistive medical technology. Specifically, it includes: deploying an intelligent interactive terminal in the user's daily living environment to continuously collect unstructured user command data through daily interactions between the user and the device; anonymizing the collected command data and extracting a multi-dimensional feature set containing command logical coherence and intent rationality; establishing a dynamic baseline model based on command data from the user's health status period, wherein the baseline model automatically expands its storage capacity according to the time dimension, forming a personalized reference database that evolves synchronously with the user's interaction characteristics; and generating a visual early warning report containing disease progression risk levels when the evolution matching degree between real-time command features and the baseline model is analyzed, and the command logical coherence and / or intent rationality continuously decline and exceed an adaptive threshold. This invention can assist in the early detection of disease risks.

Description

Technical Field

[0001] This invention relates to the field of assistive medical technology, specifically to an intelligent method and system for assessing functional impairments in Parkinson's disease. Background Technology

[0002] With the increasing aging of the population, the early diagnosis and management of chronic diseases are becoming increasingly important in the healthcare field. Parkinson's disease, as a common neurodegenerative disease, presents significant challenges to accurate diagnosis and disease monitoring due to its insidious early symptoms, slow progression, and individual variability. Timely and accurate assessment of the degree of functional impairment in Parkinson's disease patients is crucial for developing personalized treatment plans, slowing disease progression, and improving patients' quality of life.

[0003] In current technologies, the assessment of Parkinson's disease functional impairment mainly relies on the professional judgment of clinicians and limited scale assessments. Traditional assessment methods are mostly conducted in a hospital setting, requiring patients to actively cooperate in completing specific tasks or answering questionnaires, such as the Standardized Parkinson's Disease Rating Scale (UPDRS). This approach not only consumes medical resources but also, due to the limited assessment timeframes, struggles to capture the true changes in the patient's condition in daily life. Furthermore, while some sensor-based monitoring methods can acquire some physiological data, they suffer from inconvenience in wearing them and complex data interpretation, failing to achieve long-term, continuous monitoring in a natural state.

[0004] However, existing methods for assessing Parkinson's disease functional impairment ignore data from patients' daily interactions, making it impossible to dynamically track disease progression and provide early warnings, which may cause patients to miss the best treatment opportunity. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent method and system for assessing functional impairment in Parkinson's disease, and to solve the following technical problems:

[0006] Existing methods for assessing functional impairment in Parkinson's disease neglect data on interactions in patients' daily lives.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] An intelligent method for assessing functional impairment in Parkinson's disease includes the following steps:

[0009] Deploy intelligent interactive terminals in users' daily living environments to continuously collect unstructured user command data through users' daily interactions with devices;

[0010] The collected instruction data is anonymized, and a multi-dimensional feature set containing the logical coherence and rationality of the instruction is extracted to obtain real-time instruction features.

[0011] A dynamic baseline model is established based on the instruction data of the user's health status period. The baseline model automatically expands its storage capacity according to the time dimension to form a personalized reference database that evolves in sync with the user's interaction characteristics.

[0012] By analyzing the evolutionary matching degree between the real-time instruction features and the baseline model, when the deviation value of the multidimensional feature set is detected to be continuously decreasing and exceeding the adaptive threshold, a visual early warning report containing the risk level of disease progression is generated.

[0013] As a further aspect of the present invention: the collection of user instruction data specifically includes:

[0014] The user commands include voice commands and direct operation commands. The smart interactive terminal automatically activates the command acquisition function when the user performs device control, information query, or daily communication, and captures the complete interactive command content of the user. The device control includes smart home operation and electronic device parameter adjustment. The information query is a user-initiated query command. The daily communication is a device voice question and answer.

[0015] The instruction content is categorized and labeled according to the interaction scenario and stored in a local encrypted cache and a cloud database; the cloud database only retains the anonymized instruction feature vectors, and the feature vectors are desensitized through a hash algorithm.

[0016] As a further aspect of the present invention: the process of extracting a multi-dimensional feature set containing instruction logical coherence and intent rationality is as follows:

[0017] The system collects raw data generated by users through voice commands and direct operation commands, converts voice commands into text, and records the type, trigger time, and action path of operation commands. The voice text and operation commands are aligned with millisecond-level timestamps to construct a multimodal command sequence and identify multiple repetitions or contradictory expressions of user intent in the same task.

[0018] Semantic contradictions between adjacent instructions are detected and marked as short-term logical conflicts; when voice instructions and operation instructions are inconsistent in semantics or action direction, they are marked as cross-modal logical conflicts; the frequency of logical conflict events per unit time is counted, and difference weights are assigned according to the conflict type to generate a logical coherence score.

[0019] The context matching degree of the command is analyzed. Commands issued without spatiotemporal trigger conditions are marked as spatiotemporal abnormal intentions. Commands with non-coherent operation chains are marked as disordered abnormal intentions. When the voice command and operation command have no coordination relationship in the time-dense interval, they are judged as disordered abnormal intentions. Differential weights are assigned according to the type of abnormal intentions to generate an intention rationality score.

[0020] The logical coherence score and the intention rationality score are integrated into a multi-dimensional feature vector that represents the cognitive degradation of users.

[0021] As a further aspect of the present invention: the method for constructing the dynamic baseline model includes:

[0022] The initial model is based on the health status command data collected continuously when the user first registers. The initial data has been medically confirmed to have no interactive functional impairments, and the collection period covers the user's typical interactive behaviors in different time periods.

[0023] When updating the model, the newly added instruction data is aligned with the historical data over time. The data is divided into segments using a sliding window technique, and the deviation value of the multi-dimensional feature set within each time window is calculated. The deviation value is calculated based on a combination of logical coherence score and intent rationality score. When the deviation value is within a preset healthy range, the new feature is added to the baseline model and the mean of logical coherence and the fluctuation range are recalculated to ensure that the model evolves naturally with the user's normal behavior.

[0024] When the deviation value exceeds the healthy range, a manual review process is triggered. The review includes recent changes in the user's living environment, adjustments to device operating habits, or temporary emotional fluctuations. The model can only be updated after the deviation is confirmed to be caused by non-disease factors.

[0025] As a further aspect of the present invention: the calculation process of the evolution matching degree includes:

[0026] The real-time instruction features are mapped to the interactive feature space constructed by the baseline model. The feature space includes a logical coherence scoring axis, an intent rationality scoring axis, and a response timeliness coordinate axis, generating the three-dimensional coordinate position of the current feature point.

[0027] Based on the time dimension, feature points from consecutive acquisition cycles are connected to form an instruction evolution trajectory line, which reflects the trend of user interaction features changing over time.

[0028] The minimum encirclement distance between the trajectory line and the historical health trajectory cluster is calculated. The historical health trajectory cluster consists of multiple typical trajectories in the user's health state, representing the fluctuation range of normal interactive behavior. When the trajectory line continuously deviates and the deviation direction is manifested as a break in instruction logic accompanied by an increase in response latency, it is determined to be an abnormal degradation mode. The logic break is quantified by the frequency of instruction context jumps, and the response latency is calculated by the time difference of the device executing the instruction.

[0029] As a further aspect of the present invention: the method for setting the adaptive threshold includes:

[0030] Based on the historical degradation pattern of logical coherence scores and the fluctuation trend of intention rationality scores in the baseline model, a dynamic threshold curve is established. The shape of the curve is jointly determined by the rate of decline of logical coherence scores and the abnormal fluctuation amplitude of intention rationality scores. Corresponding weight coefficients are set for logical coherence scores and intention rationality scores.

[0031] When the rate of decline of the logical coherence score exceeds the current threshold curve, a fuzzy logic algorithm is used to calculate the comprehensive risk index. The input variables of the algorithm include the logical break strength, the frequency of abnormal fluctuations in the rationality of intent, and the frequency of cross-modal conflict. The output risk level is mapped to the [0,1] interval. If the index value exceeds the set critical value, a high-risk warning is triggered.

[0032] As a further aspect of the present invention: the method for generating the visualized early warning report includes:

[0033] The system overlays and compares the instruction evolution trajectory with historical health trajectory clusters, using a dynamic color heatmap to differentiate risk levels. The report marks the interaction feature type with the fastest degradation rate, including the frequency of logical breakage and intent fluctuation, and links it to historical behavior records in the user's health profile to compare the consistency between the current degradation trend and previous medical diagnosis results. The visualized early warning report includes key degradation time nodes. When a key degradation time node is clicked, the specific instruction data and execution results of the corresponding time node are traced back to help locate the source of abnormal behavior.

[0034] As a further aspect of the present invention: all instruction data undergoes feature extraction and anonymization locally. The extraction process employs edge computing technology to ensure that the original instruction data does not leave the user terminal. The anonymized feature vectors are uploaded to a cloud database using an asymmetric encryption algorithm. The database adopts an independent partition storage architecture, where user identity information and interactive data are physically isolated, and the access permissions for each partition are controlled by an independent key.

[0035] Medical personnel need to access the full report through dual verification using a user authorization code and a system dynamic key. The dynamic key is valid for a single session and expires automatically after the session ends. After the data retention period ends, the instruction data stored in the cloud and locally will be automatically destroyed.

[0036] As a further aspect of the present invention, a verification mechanism is also included.

[0037] After a high-risk warning report is generated, the system automatically extracts the user's instruction data in various interactive scenarios within a set time period, including device control, complex task execution, and multi-turn dialogue interaction; compares the distribution of logical coherence scores, the decay curve of intent rationality scores, and response stability in different scenarios. If all scenarios show a consistent degradation trend, it is judged as a risk of disease progression.

[0038] If anomalies occur only in specific scenarios, environmental interference factors are identified, including device response latency and network connection stability. If environmental interference is the primary cause, the baseline model parameters are recalibrated and false alarm events are flagged.

[0039] The present invention also includes an intelligent Parkinson's disease functional impairment assessment system for implementing the above-described intelligent Parkinson's disease functional impairment assessment method, comprising:

[0040] The data acquisition module deploys intelligent interactive terminals in the user's daily living environment and continuously collects unstructured user command data through the user's daily interaction with the device;

[0041] The instruction processing module is used to anonymize the collected instruction data and extract a multi-dimensional feature set that includes the logical coherence and rationality of the instruction intent.

[0042] The model building module establishes a dynamic baseline model based on instruction data during the user's health status period. The baseline model automatically expands its storage capacity according to the time dimension, forming a personalized reference database that evolves synchronously with the user's interaction characteristics.

[0043] The cognitive analysis module is used to analyze the evolutionary matching degree between real-time command features and baseline models. When the deviation value of the multidimensional feature set is detected to be continuously decreasing and exceeds the adaptive threshold, a visual early warning report containing the risk level of disease progression is generated.

[0044] The beneficial effects of this invention are:

[0045] This invention utilizes an intelligent interactive terminal deployed in the user's daily living environment to comprehensively collect user command data, including voice and direct operation data, thus solving the problem of limited data collection in existing technologies. It anonymizes the command data and extracts a multi-dimensional feature set containing logical coherence and reasonable intent, enabling more accurate and objective analysis of the user's functional status compared to traditional subjective judgment. A dynamic baseline model is established based on health status command data, which evolves naturally with the user's normal behavior, achieving real-time dynamic assessment. By calculating the evolution matching degree and setting adaptive thresholds to generate visual early warning reports, it can promptly detect the risk of disease progression, overcoming the shortcomings of existing methods in timely monitoring of subtle disease progression, achieving early warning, and preventing patients from missing the optimal treatment opportunity. Attached Figure Description

[0046] The invention will now be further described with reference to the accompanying drawings.

[0047] Figure 1 This is a flowchart illustrating an intelligent method for assessing functional impairment in Parkinson's disease according to the present invention. Detailed Implementation

[0048] 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.

[0049] Please see Figure 1 As shown, this invention is an intelligent method for assessing functional impairment in Parkinson's disease, comprising the following steps:

[0050] Data Acquisition: Intelligent interactive terminals are meticulously deployed within users' daily living environments. These terminals automatically and continuously collect unstructured user command data as users perform various operations on their devices. Both voice commands and direct operational instructions are accurately captured.

[0051] Data Processing and Feature Extraction: After collecting the instruction data, it is immediately anonymized. Algorithms are used to deeply extract a multi-dimensional feature set containing the logical coherence and rationality of the instruction's intent, carefully identifying the logical relationships and rationality of the intent within the instruction to obtain real-time instruction features.

[0052] Dynamic baseline model construction: Based on instruction data collected during periods when the user is in a healthy state, a dynamic baseline model is constructed. This baseline model has unique advantages, as it can automatically expand its storage capacity according to the time dimension, closely aligning with the user's daily interaction habits, and gradually forming a personalized reference database that evolves in sync with the user's interaction characteristics.

[0053] Assessment and Early Warning: By deeply analyzing the evolutionary matching degree between real-time instruction characteristics and baseline models, once the logical coherence score is detected to be continuously declining and the intention rationality score is abnormally fluctuating beyond the adaptive threshold, the system will quickly generate a visual early warning report containing the risk level of disease progression, providing an intuitive and crucial reference for the assessment of Parkinson's disease functional impairment.

[0054] In a preferred embodiment of the present invention, the user instruction data is collected as follows:

[0055] User commands primarily fall into two categories: voice commands and direct operation commands. The intelligent interactive terminal possesses a highly intelligent command acquisition and activation mechanism. When users engage in device control, such as operating smart home devices to easily adjust light brightness and temperature, or adjusting electronic device parameters like TV volume or computer display resolution; when querying information, such as asking about weather or news; or when conducting daily conversations through voice Q&A, chatting about everyday matters, or discussing hobbies, the terminal automatically and sensitively activates its command acquisition function, accurately capturing the complete user interaction commands to ensure no crucial information is missed. After acquiring the commands, they are meticulously categorized and labeled according to the interaction scenario. For example, smart home operation commands are labeled "home control," and information query commands are labeled "information retrieval." These labeled commands are simultaneously stored in a local encrypted cache and a cloud database. The local encrypted cache ensures data immediacy and security, preventing data loss before transmission. The cloud database, on the other hand, only retains the instruction feature vectors after anonymization, and these feature vectors are de-identified using advanced hash algorithms, further strengthening the defenses for data privacy and avoiding the risk of user information leakage.

[0056] In another preferred embodiment of the present invention, the process of extracting a multi-dimensional feature set that includes instruction logical coherence and intent rationality is as follows:

[0057] First, comprehensive raw data generated by users through voice commands and direct operation commands is collected. For voice commands, efficient and accurate speech recognition technology is used for text conversion, accurately transforming the user's spoken words into text form. For operation commands, their type is recorded in detail, such as whether it is a click, swipe, or long press; the trigger time is precisely recorded to the millisecond level for subsequent time series analysis; and the complete action path is recorded, such as the trajectory of swiping from point A to point B on the screen. Next, the voice text and operation commands are strictly aligned with millisecond-level timestamps to construct a multimodal command sequence. This sequence can accurately identify multiple repetitions or contradictory expressions of user intent within the same task. For example, if a user sets an alarm clock, first saying "set for 7 AM" and then manually setting it to 8 AM, this contradictory expression can be accurately captured. Next, semantic contradictions between adjacent commands are deeply detected, and once detected, they are immediately marked as short-term logical conflicts. When voice commands and operation commands are inconsistent in semantics or action direction, such as saying "turn on the TV" but performing the action of turning off the TV, it is marked as a cross-modal logical conflict. Then, the frequency of logical conflict events per unit time is statistically analyzed. Different types of conflicts are assigned different weights based on their varying impact on the evaluation. A logical coherence score is generated through scientific calculation to quantify the logical coherence between instructions. Regarding intent rationality analysis, the contextual matching degree of the instructions is carefully analyzed. Instructions issued without spatiotemporal triggers, such as suddenly saying "go on a trip tomorrow" without any time-related prompts, are marked as spatiotemporally abnormal intents. Instructions with disjointed operation chains, such as clicking different function modules illogically while operating a computer, are marked as disordered abnormal intents. When voice instructions and operation instructions have no synergistic relationship within a time-intensive period, such as asking for a recipe while randomly swiping images on a phone, they are judged as disordered abnormal intents. Similarly, different weights are assigned according to the type of abnormal intent to generate an intent rationality score. Finally, the logical coherence score and intent rationality score are cleverly fused using a specific fusion algorithm to create a multi-dimensional feature vector representing user cognitive degradation, providing crucial data support for subsequent evaluations.

[0058] In another preferred embodiment of the present invention, the method for constructing the dynamic baseline model is as follows:

[0059] The initial model of the dynamic baseline model. To ensure the reliability and representativeness of the initial data, several key points need to be met.

[0060] First, this initial data must undergo rigorous verification by medical professionals to ensure that the user does not experience any functional impairments during data collection. Medical personnel will use a variety of medical examination methods and their professional knowledge to comprehensively assess the user's physical condition and cognitive function. Only after confirming that the user is in good health and can interact normally with the device will the collected data be incorporated into the initial model.

[0061] Secondly, the data collection period must be sufficiently broad and comprehensive to cover typical user interactions across different time periods. For example, it should consider users' different lifestyles and interaction habits on weekdays and weekends, as well as their different activity patterns during the day and night. This means that data collection may need to continue for weeks or even months to ensure that all possible user interaction scenarios and behavioral patterns are captured.

[0062] Over time, user interactions may naturally change. To ensure the dynamic baseline model accurately reflects normal user behavior, it needs to be updated periodically. The update process mainly includes the following steps:

[0063] Time series alignment: When new instruction data is added, it must first be aligned with historical data over time. This step ensures that data from different time periods accurately correspond in the time dimension during subsequent analysis, facilitating reasonable comparisons and calculations. For example, new and historical data can be arranged chronologically using timestamps and other information, ensuring accurate matching at each point in time.

[0064] Data Segmentation and Deviation Calculation: A sliding window technique is used to segment the aligned data into fixed-length segments. Each segment represents user interaction data within a specific time period. Then, for the data within each time window, a deviation value for the multi-dimensional feature set is calculated. This deviation value is based on a combined calculation of logical coherence and intent rationality scores. The logical coherence score reflects the rationality of the logical relationships between user commands, while the intent rationality score measures whether the intent of the user commands is reasonable. By comprehensively considering these two scores, the characteristics of user interaction behavior can be evaluated more comprehensively.

[0065] Deviation Value Judgment and Model Update: A preset health range is established to determine whether the currently calculated deviation value is within the normal range. If the deviation value is within the preset health range, it indicates that the user's interaction behavior has not undergone significant abnormal changes. At this time, new features can be added to the baseline model, and the logical coherence mean and fluctuation range can be recalculated. This ensures that the model can be continuously updated as the user's normal behavior evolves naturally, maintaining its adaptability to the user's current behavior patterns.

[0066] When the deviation value exceeds the healthy range, it indicates a possible abnormal change in the user's interaction behavior, triggering a manual review process. This manual review involves a detailed investigation of factors such as recent changes in the user's living environment, adjustments to device operating habits, or temporary emotional fluctuations. For example, a user might have changed their living environment due to moving, leading to a change in their interaction with smart home devices; or they might have learned new device operating skills, thus changing their operating habits; or the user might have experienced emotional fluctuations due to recent work stress, affecting their interaction with the devices. Only after confirming that these non-disease factors are the cause of the deviation is the model updated to avoid errors caused by abnormal data due to non-disease factors influencing the model.

[0067] In another preferred embodiment of the present invention, the calculation process of the evolution matching degree is as follows:

[0068] The real-time acquired instruction features are mapped to an interaction feature space constructed from the baseline model. This interaction feature space is a three-dimensional space containing a logical coherence scoring axis, an intent rationality scoring axis, and a response timeliness coordinate axis. By analyzing and calculating the real-time instruction features, scores are obtained in the three aspects of logical coherence, intent rationality, and response timeliness. These scores are then used as coordinate values ​​to generate the three-dimensional coordinate position of the current feature point in the interaction feature space. For example, assuming the real-time instruction has a logical coherence score of 80, an intent rationality score of 75, and a response timeliness score of 85, then the three-dimensional coordinates of the feature point are (80, 75, 85).

[0069] Based on the time dimension, feature points within a continuous data collection period are connected sequentially to form an instruction evolution trajectory line. This trajectory line can intuitively reflect the trend of user interaction characteristics changing over time. For example, if a user's logical coherence score gradually decreases over a period of time, the trajectory line will show a downward trend on the logical coherence score axis; if the intent rationality score fluctuates significantly, the trajectory line will show a more tortuous shape on the intent rationality score axis. By observing the shape and direction of the trajectory line, a preliminary judgment can be made on changes in user interaction behavior.

[0070] To more accurately assess the difference between the current trajectory line and the user's interaction behavior under historical health conditions, it is necessary to calculate the minimum enclosing distance between the current trajectory line and the historical healthy trajectory cluster. The historical healthy trajectory cluster consists of multiple typical trajectories of the user under healthy conditions; these typical trajectories represent the fluctuation range of the user's normal interaction behavior. By calculating the minimum enclosing distance between the current trajectory line and the historical healthy trajectory cluster, the degree of deviation of the current trajectory line from the normal fluctuation range can be quantified. For example, some geometric algorithms can be used to calculate the shortest distance between the trajectory line and the trajectory cluster. The smaller this distance, the closer the current trajectory line is to the historical healthy trajectory, and the more normal the user's interaction behavior; conversely, the larger the distance, the more likely the user's interaction behavior has changed abnormally.

[0071] When the trajectory line continuously deviates from the historical healthy trajectory cluster, and the deviation is manifested as logical breaks in instructions accompanied by an increase in response latency, it is identified as an abnormal degradation mode. Logical breaks can be quantified by the frequency of instruction context jumps, that is, counting the number of times the contextual relationship between instructions jumps or becomes disjointed within a certain time period. The response latency rate is calculated by the time difference between the user issuing an instruction and the device actually executing the instruction, that is, recording the time interval between the user issuing an instruction and the device actually executing the instruction, and calculating the average latency time. If the frequency of logical breaks is high, and the response latency rate also increases significantly, it indicates that the user's interaction ability may have deteriorated, and there is a risk of Parkinson's disease-related functional impairment.

[0072] In another preferred embodiment of the present invention, the method for setting the adaptive threshold is as follows:

[0073] Based on the historical degradation patterns of logical coherence scores and the fluctuation trends of intention rationality scores in the baseline model, a dynamic threshold curve is established. The shape of this curve is jointly determined by the rate of decline of logical coherence scores and the magnitude of abnormal fluctuations in intention rationality scores. Specifically, through analysis of historical data, the rate of decline of logical coherence scores and the magnitude of abnormal fluctuations in intention rationality scores within different time periods are statistically analyzed, and then a curve is fitted based on these statistical data. For example, if logical coherence scores decline rapidly while intention rationality scores fluctuate abnormally significantly within certain time periods, the threshold on the curve corresponding to these time periods will be increased accordingly to adapt to this change.

[0074] To more accurately reflect the importance of logical coherence and intention rationality scores in risk assessment, corresponding weighting coefficients need to be assigned to these two scores. The determination of these weighting coefficients can be based on medical research and practical experience. For example, if extensive experiments and analyses reveal that logical coherence scores are more critical in assessing Parkinson's disease functional impairment, then a higher weighting coefficient can be assigned to the logical coherence score.

[0075] When the rate of decline in the logical coherence score exceeds the current threshold curve, it indicates that the user's interaction behavior may have undergone abnormal changes. At this point, a fuzzy logic algorithm is used to calculate the comprehensive risk index. The fuzzy logic algorithm is capable of handling uncertain and fuzzy information. Its input variables include the strength of logical breaks, the frequency of abnormal fluctuations in intent rationality, and the frequency of cross-modal conflicts. Through fuzzification, rule-based reasoning, and defuzzification of these input variables, a risk level is output and mapped to the [0,1] interval. For example, a risk level of 0 indicates no risk, and a risk level of 1 indicates extremely high risk. A threshold is set; when the calculated comprehensive risk index value exceeds this threshold, the system triggers a high-risk warning. This means that the system believes the user has a high risk of Parkinson's disease progression and requires timely further examination and intervention.

[0076] In another preferred embodiment of the present invention, the method for generating a visual early warning report is as follows:

[0077] By precisely overlaying and comparing the evolution trajectory of user commands with historical health trajectory clusters, changes in user interaction behavior are immediately apparent. To more clearly differentiate risk levels, the system employs dynamic color heatmap technology, with different risk levels presented visually in different colors. For example, low-risk areas are displayed in green, medium-risk areas in yellow, and high-risk areas in red, allowing users to instantly identify the current risk level. The report also details the interaction characteristics with the fastest degradation rate, which may include logic break frequency (the number of times the command logic breaks per unit time), response latency (device response latency time accurate to the millisecond), or execution error rate (the percentage of erroneous commands executed out of the total number of commands executed). Simultaneously, the report intelligently links to historical behavior records in the user's health profile, meticulously comparing the current degradation trend with previous medical diagnoses to check for consistency. The report specifically includes key degradation time nodes. When users click on these key degradation time nodes, the system quickly retrieves the specific command content and execution results for that period, accurately assisting in locating the source of abnormal behavior and providing strong data support for subsequent in-depth analysis of the condition and the development of intervention measures.

[0078] In another preferred embodiment of the invention, all instruction data undergoes a comprehensive and efficient feature extraction and anonymization process locally from the very beginning of acquisition. During the feature extraction stage, cutting-edge edge computing technology is employed, utilizing the computing power of the user terminal device itself to perform deep mining of the original instruction data locally, extracting features containing key information such as the logical coherence and rationality of the intent. This local processing method ensures that the original instruction data does not leave the user terminal to the greatest extent possible, eliminating the risk of data leakage during transmission. After feature extraction, the data is immediately anonymized. Through complex and reliable algorithms, all sensitive information that could identify the user is removed, leaving only the necessary features for Parkinson's disease functional impairment assessment. The anonymized feature vectors are then uploaded to a cloud database using an asymmetric encryption algorithm. The asymmetric encryption algorithm acts like an impenetrable "digital armor" for the data, using different keys for encryption and decryption, greatly enhancing the security of data transmission. The cloud database employs a carefully designed independent partition storage architecture, with each user's interaction data stored separately in an independent partition. More importantly, user identity information and interaction data are physically isolated, like placing them in different "safes," further ensuring data security. Moreover, access permissions for each partition are strictly controlled by an independent key; only authorized personnel holding the correct key can access data in a specific partition. This granular access management model effectively prevents unauthorized access and misuse of data.

[0079] Regarding data access permissions, medical providers must undergo a rigorous dual verification process using both a user authorization code and a system dynamic key to obtain complete reports. The user authorization code is an access permission credential actively granted by the user based on trust in the medical provider, ensuring the legality of data access and user autonomy. The system dynamic key generates a unique temporary key for each access, with its validity strictly limited to a single session. Once it expires, the dynamic key automatically becomes invalid, acting like a time-limited "digital lock." Even if the key information is inadvertently leaked, it cannot be used to access data after the expiration date, adding another robust layer of defense for data security. When the data retention period ends, both cloud-stored instruction data and locally cached related data are automatically triggered through an irreversible destruction process. This process uses complex algorithms to repeatedly overwrite and erase the data, ensuring that the data cannot be recovered and completely eliminating the potential risks associated with data retention.

[0080] In another preferred embodiment of the present invention, the verification mechanism includes:

[0081] After generating a high-risk warning report, a thorough verification process is immediately and automatically initiated. The system comprehensively extracts instruction data generated by the user in various interactive scenarios within a set time period. These interactive scenarios cover device control, such as users turning smart home devices on and off or adjusting them; complex task execution, such as using electronic devices to complete a series of logically related tasks, such as product search, order placement, and payment in an online shopping process; and multi-turn dialogue interaction, such as engaging in continuous question-and-answer exchanges with a smart voice assistant. After extracting the data, the system meticulously compares the logical coherence of instructions in different scenarios, judging whether the semantic and logical connections between instructions are tight; execution accuracy, statistically analyzing the proportion of correctly executed instructions; and response stability, checking whether the device responds to instructions stably and smoothly, without any delays or abnormal interruptions. If a consistent degradation trend is observed in all these different scenarios, such as deteriorating logical coherence of instructions, reduced execution accuracy, and decreased response stability, the system will cautiously determine it as a risk of disease progression, providing strong evidence for medical decision-making.

[0082] However, the system will not jump to conclusions if anomalies only occur in specific scenarios. Instead, it will quickly identify potential environmental interference factors, including device response latency (excessive time from receiving a command to starting execution, potentially due to device performance issues or handling too many tasks simultaneously) and network connection stability (poor signal strength, frequent interruptions, or high latency can also affect user interaction). If analysis determines environmental interference is the primary cause, the system will promptly recalibrate the baseline model parameters, optimizing the model to better reflect the actual environment. Simultaneously, the system will rigorously flag false alarms and record detailed information about the anomaly for subsequent analysis and summarization, continuously improving the accuracy and reliability of the evaluation system.

[0083] The present invention also includes an intelligent Parkinson's disease functional impairment assessment system for implementing the above-described intelligent Parkinson's disease functional impairment assessment method, comprising:

[0084] The data acquisition module deploys intelligent interactive terminals in the user's daily living environment and continuously collects unstructured user command data through the user's daily interaction with the device;

[0085] The instruction processing module is used to anonymize the collected instruction data and extract a multi-dimensional feature set that includes the logical coherence and rationality of the instruction intent.

[0086] The model building module establishes a dynamic baseline model based on instruction data during the user's health status period. The baseline model automatically expands its storage capacity according to the time dimension, forming a personalized reference database that evolves synchronously with the user's interaction characteristics.

[0087] The cognitive analysis module is used to analyze the evolutionary matching degree between real-time command features and baseline models. When the deviation value of the multidimensional feature set is detected to be continuously decreasing and exceeds the adaptive threshold, a visual early warning report containing the risk level of disease progression is generated.

[0088] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. An intelligent method for assessing functional impairment in Parkinson's disease, characterized in that, Includes the following steps: Deploy intelligent interactive terminals in users' daily living environments to continuously collect unstructured user command data through users' daily interactions with devices; The collected instruction data is anonymized, and a multi-dimensional feature set containing the logical coherence and rationality of the instruction is extracted to obtain real-time instruction features. A dynamic baseline model is established based on the instruction data of the user's health status period. The baseline model automatically expands its storage capacity according to the time dimension to form a personalized reference database that evolves in sync with the user's interaction characteristics. By analyzing the evolutionary matching degree between the real-time instruction features and the baseline model, when the deviation value of the multidimensional feature set is detected to be continuously decreasing and exceeding the adaptive threshold, a visual early warning report containing the risk level of disease progression is generated. The process of extracting a multi-dimensional feature set that includes the logical coherence of instructions and the rationality of intent is as follows: The user instructions include voice instructions and direct operation instructions. The raw data generated by the user through voice instructions and direct operation instructions are collected. The voice instructions are converted into text, and the operation instructions are recorded in terms of type, trigger time and action path. The voice text and operation instructions are aligned with millisecond-level timestamps to construct a multimodal instruction sequence and identify multiple repetitions or contradictory expressions of user intent in the same task. Semantic contradictions between adjacent instructions are detected and marked as short-term logical conflicts; When voice commands and operation commands are inconsistent in semantics or action direction, they are marked as cross-modal logic conflicts. The frequency of logical conflict events per unit time is statistically analyzed, and differential weights are assigned according to the conflict type to generate a logical coherence score. The context matching degree of the command is analyzed. Commands issued without spatiotemporal trigger conditions are marked as spatiotemporal abnormal intentions. Commands with non-coherent operation chains are marked as disordered abnormal intentions. When the voice command and operation command have no coordination relationship in the time-dense interval, they are judged as disordered abnormal intentions. Differential weights are assigned according to the type of abnormal intentions to generate an intention rationality score. The logical coherence score and the intention rationality score are merged into a multi-dimensional feature vector that represents the cognitive degradation of users; The calculation process for the evolution matching degree includes: The real-time instruction features are mapped to the interactive feature space constructed by the baseline model. The feature space includes a logical coherence scoring axis, an intent rationality scoring axis, and a response timeliness coordinate axis, generating the three-dimensional coordinate position of the current feature point. Based on the time dimension, feature points from consecutive acquisition cycles are connected to form an instruction evolution trajectory line, which reflects the trend of user interaction features changing over time. The minimum encirclement distance between the trajectory line and the historical health trajectory cluster is calculated. The historical health trajectory cluster consists of multiple typical trajectories in the user's health state, representing the fluctuation range of normal interactive behavior. When the trajectory line continuously deviates and the deviation direction is manifested as a break in instruction logic accompanied by an increase in response latency, it is determined to be an abnormal degradation mode. The logic break is quantified by the frequency of instruction context jumps, and the response latency is calculated by the time difference of the device executing the instruction.

2. The intelligent method for assessing functional impairment in Parkinson's disease according to claim 1, characterized in that, The collection of user command data specifically includes: The user commands include voice commands and direct operation commands. The smart interactive terminal automatically activates the command acquisition function when the user performs device control, information query, or daily communication, and captures the complete interactive command content of the user. The device control includes smart home operation and electronic device parameter adjustment. The information query is a user-initiated query command. The daily communication is a device voice question and answer. The instruction content is categorized and labeled according to the interaction scenario and stored in a local encrypted cache and a cloud database; the cloud database only retains the anonymized instruction feature vectors, and the feature vectors are desensitized through a hash algorithm.

3. The intelligent method for assessing functional impairment in Parkinson's disease according to claim 1, characterized in that, The method for constructing the dynamic baseline model includes: The initial model is based on the health status command data collected continuously when the user first registers. The initial data has been medically confirmed to have no interactive functional impairments, and the collection period covers the user's typical interactive behaviors in different time periods. When updating the model, the newly added instruction data is aligned with the historical data over time. The data is divided into segments using a sliding window technique, and the deviation value of the multi-dimensional feature set within each time window is calculated. The deviation value is calculated based on a combination of logical coherence score and intent rationality score. When the deviation value is within a preset healthy range, the new feature is added to the baseline model and the mean of logical coherence and the fluctuation range are recalculated to ensure that the model evolves naturally with the user's normal behavior. When the deviation value exceeds the healthy range, a manual review process is triggered. The review includes recent changes in the user's living environment, adjustments to device operating habits, or temporary emotional fluctuations. The model can only be updated after the deviation is confirmed to be caused by non-disease factors.

4. The intelligent method for assessing functional impairment in Parkinson's disease according to claim 1, characterized in that, The method for setting the adaptive threshold includes: Based on the historical degradation pattern of logical coherence scores and the fluctuation trend of intention rationality scores in the baseline model, a dynamic threshold curve is established. The shape of the curve is jointly determined by the rate of decline of logical coherence scores and the abnormal fluctuation amplitude of intention rationality scores. Corresponding weight coefficients are set for logical coherence scores and intention rationality scores. When the rate of decline of the logical coherence score exceeds the current threshold curve, a fuzzy logic algorithm is used to calculate the comprehensive risk index. The input variables of the algorithm include the logical break strength, the frequency of abnormal fluctuations in the rationality of intent, and the frequency of cross-modal conflict. The output risk level is mapped to the [0,1] interval. If the index value exceeds the set critical value, a high-risk warning is triggered.

5. The intelligent method for assessing functional impairment in Parkinson's disease according to claim 1, characterized in that, The method for generating the visualized early warning report includes: The system overlays and compares the instruction evolution trajectory with historical health trajectory clusters, using a dynamic color heatmap to differentiate risk levels. The report marks the interaction feature type with the fastest degradation rate, including the frequency of logical breakage and intent fluctuation, and links it to historical behavior records in the user's health profile to compare the consistency between the current degradation trend and previous medical diagnosis results. The visualized early warning report includes key degradation time nodes. When a key degradation time node is clicked, the specific instruction data and execution results of the corresponding time node are traced back to help locate the source of abnormal behavior.

6. The intelligent method for assessing functional impairment in Parkinson's disease according to claim 1, characterized in that, All instruction data undergoes feature extraction and anonymization locally. The extraction process employs edge computing technology to ensure that the original instruction data does not leave the user terminal. The anonymized feature vectors are uploaded to a cloud database using an asymmetric encryption algorithm. The database adopts an independent partition storage architecture, physically isolating user identity information from interaction data, and the access permissions for each partition are controlled by an independent key. Medical personnel need to access the full report through dual verification using a user authorization code and a system dynamic key. The dynamic key is valid for a single session and expires automatically after the session ends. After the data retention period ends, the instruction data stored in the cloud and locally will be automatically destroyed.

7. The intelligent method for assessing functional impairment in Parkinson's disease according to claim 1, characterized in that, It also includes a verification mechanism: After a high-risk warning report is generated, the system automatically extracts the user's instruction data in various interactive scenarios within a set time period, including device control, complex task execution, and multi-turn dialogue interaction; compares the distribution of logical coherence scores, the decay curve of intent rationality scores, and response stability in different scenarios. If all scenarios show a consistent degradation trend, it is judged as a risk of disease progression. If anomalies only occur in specific scenarios, environmental interference factors are identified, including device response latency and network connection stability. If environmental interference is the primary cause, then recalibrate the baseline model parameters and label false alarm events.

8. An intelligent Parkinson's disease functional impairment assessment system, used to implement the intelligent Parkinson's disease functional impairment assessment method according to any one of claims 1-7, characterized in that, include: The data acquisition module deploys intelligent interactive terminals in the user's daily living environment and continuously collects unstructured user command data through the user's daily interaction with the device; The instruction processing module is used to anonymize the collected instruction data and extract a multi-dimensional feature set that includes the logical coherence and rationality of the instruction intent. The model building module establishes a dynamic baseline model based on instruction data during the user's health status period. The baseline model automatically expands its storage capacity according to the time dimension, forming a personalized reference database that evolves synchronously with the user's interaction characteristics. The cognitive analysis module is used to analyze the evolutionary matching degree between real-time command features and baseline models. When the deviation value of the multidimensional feature set is detected to be continuously decreasing and exceeds the adaptive threshold, a visual early warning report containing the risk level of disease progression is generated.

Images