Intelligent Parkinson's disease dysfunction assessment method and system

Abstract

The invention discloses an intelligent Parkinson's disease dysfunction assessment method and system, and belongs to the technical field of medical assistance, and the method specifically comprises the steps: deploying an intelligent interaction terminal in a daily living environment of a user, and continuously collecting unstructured user instruction data through the daily interaction behavior of the user and equipment; carrying out anonymization processing on the collected instruction data, and extracting a multi-dimensional feature set containing instruction logic continuity and intention rationality; a dynamic baseline model is established based on the instruction data of the user in the health state period, the baseline model automatically expands the storage capacity according to the time dimension, and a personalized reference database synchronously evolved with user interaction characteristics is formed; by analyzing the evolution matching degree between real-time instruction features and a baseline model, when it is detected that instruction logic continuity or / and intention rationality continuously declines and exceeds a self-adaptive threshold value, a visual early warning report containing a disease progress risk level is generated; the method can assist in finding early risks of diseases.

Description

Technical Field

[0001] The present invention relates to the field of medical assistance technologies, and particularly to an intelligent method and system for evaluating Parkinson's disease dysfunction. Background Art

[0002] With the aggravation of population aging, the early diagnosis and management of chronic diseases have become increasingly important in the field of medical health. Parkinson's disease, as a common neurodegenerative disease, has insidious early symptoms, slow disease progression and individual differences, posing great challenges to accurate diagnosis and disease monitoring. Timely and accurately evaluating the degree of dysfunction in Parkinson's disease patients is crucial for formulating personalized treatment plans, delaying disease progression and improving the quality of life of patients. In the prior art, the evaluation of Parkinson's disease dysfunction mainly relies on the professional judgment of clinicians and limited scale evaluations. Traditional evaluation methods are mostly carried out in a hospital environment, requiring patients to actively cooperate to complete specific tasks or answer questionnaires, such as the Unified Parkinson's Disease Rating Scale (UPDRS). This method not only consumes medical resources, but also, due to limited evaluation time points, it is difficult to capture the real disease changes of patients in their daily lives. In addition, some sensor-based monitoring means, although they can obtain some physiological data, have problems such as inconvenient wearing and complex data interpretation, and cannot achieve continuous monitoring in a long-term and natural state. However, the existing methods for evaluating Parkinson's disease dysfunction ignore the interaction data of patients in their daily lives, and cannot dynamically track and early warn of disease progression, resulting in patients possibly missing the best treatment opportunity. Summary of the Invention

[0003] The purpose of the present invention is to provide an intelligent method and system for evaluating Parkinson's disease dysfunction, and solve the following technical problems: The existing methods for evaluating Parkinson's disease dysfunction ignore the interaction data of patients in their daily lives.

[0004] The purpose of the present invention can be achieved through the following technical solutions: An intelligent method for evaluating Parkinson's disease dysfunction, comprising the following steps: Deploy intelligent interaction terminals in the daily living environment of users, and continuously collect unstructured user instruction data through the daily interaction behaviors between users and the devices; Anonymize the collected instruction data, and extract a multi-dimensional feature set including the logical coherence and intention rationality of the instructions; Establish a dynamic baseline model based on the instruction data in the user's healthy state period, and the baseline model automatically expands the storage capacity according to the time dimension to form a personalized reference database that synchronously evolves with the user's interaction characteristics; By analyzing the evolutionary match between real-time instruction features and the baseline model, when it is detected that the deviation value of the multidimensional feature set continues to decrease and exceeds the adaptive threshold, a visual early warning report containing the risk level of disease progression is generated.

[0005] As a further solution of the present invention: the collection of the user instruction data specifically includes: The user instructions include voice instructions and direct operation instructions. When the user performs device control, information query or daily communication, the intelligent interactive terminal automatically activates the instruction collection function and intercepts the complete interactive instruction content of the user; the device control includes smart home operation and electronic equipment parameter adjustment, the information query is the user's active query instruction, and the daily communication is the device voice question and answer; The command content is classified and labeled according to the interaction scenario and stored in the local encrypted cache area and cloud database; the cloud database only retains the anonymized command feature vector, and the feature vector is desensitized through a hash algorithm.

[0006] As a further solution of the present invention: the process of extracting a multi-dimensional feature set including instruction logic coherence and intention rationality is: Collect the raw data generated by users through voice commands and direct operation commands, convert voice commands into text, and record the type, trigger time and action path of operation commands; align voice text and operation commands by millisecond timestamps, construct multimodal command sequences, and identify multiple repetitions or contradictory expressions of user intentions in the same task; Detect semantic contradictions between adjacent instructions and mark them as short-term logic conflicts; when voice instructions and operation instructions are inconsistent in semantics or action direction, mark them as cross-modal logic conflicts; count the frequency of logic conflict events per unit time, and assign difference weights according to the conflict type to generate a logic coherence score; Analyze the context matching of the instructions, mark the instructions issued without spatiotemporal trigger conditions as spatiotemporal abnormal intentions, mark the instructions of incoherent operation chains as disordered abnormal intentions, and judge them as disordered abnormal intentions when the voice instructions and operation instructions have no synergistic relationship in the time-intensive interval. Assign difference weights according to the types of abnormal intentions to generate intention rationality scores; The logical coherence score and intention rationality score are fused into a multidimensional feature vector to characterize user cognitive degradation.

[0007] As a further solution of the present invention: the method for constructing the dynamic baseline model includes: The initial model uses the health status instruction data collected continuously when the user first registers. The initial data has been confirmed by medical staff to have no interactive function impairment, and the collection cycle covers the user's typical interactive behaviors in different time periods. When the model is updated, the newly added instruction data is aligned with the historical data in a time series. Data segments are divided through the sliding window technique, and the deviation value of the multi-dimensional feature set within each time window is calculated. The deviation value is comprehensively calculated based on the logical coherence score and the intention rationality score. When the deviation value is within the preset healthy range, the new features are added to the baseline model, and the logical coherence mean value and the fluctuation range are recalculated to ensure the natural evolution of the model with the normal behavior of the user. When the deviation value exceeds the healthy range, an artificial review process is triggered. The review content includes changes in the user's recent living environment, adjustments to device operation habits, or temporary mood swings. The model update is only allowed after confirming that the deviation is not caused by disease factors.

[0008] As a further solution of the present invention: The calculation process of the evolution matching degree includes: Map the real-time instruction features into the interaction feature space constructed by the baseline model. The feature space includes a logical coherence score axis, an intention rationality score axis, and a response timeliness coordinate axis, and generate the three-dimensional coordinate position of the current feature point. Connect the feature points of consecutive acquisition cycles based on the time dimension to form an instruction evolution trajectory line. The trajectory line reflects the trend of the user's interaction features changing over time. Calculate the minimum enclosing distance between the trajectory line and the historical healthy trajectory cluster. The historical healthy trajectory cluster is composed of multiple typical trajectories in the user's healthy state, representing the fluctuation range of normal interaction behaviors. When the trajectory line continuously deviates and the deviation direction shows an instruction logic break accompanied by an increase in the response delay rate, it is determined as an abnormal degradation mode. The logic break is quantified by the instruction context jump frequency, and the response delay rate is calculated by the time difference of the device executing the instruction.

[0009] As a further solution of the present invention: The setting method of the adaptive threshold includes: According to the historical degradation law of the logical coherence score and the fluctuation trend of the intention rationality score in the baseline model, establish a dynamic threshold curve. The curve form is jointly determined by the decline rate of the logical coherence score and the abnormal fluctuation amplitude of the intention rationality score. Set corresponding weight coefficients for the logical coherence score and the intention rationality score. When the decline rate of the logical coherence score exceeds the current threshold curve, use the fuzzy logic algorithm to calculate the comprehensive risk index. The input variables of the algorithm include the logical break strength, the abnormal fluctuation frequency of the intention rationality, and the cross-modal conflict frequency. The output risk level is mapped to the [0,1] interval. If the index value breaks through the set critical value, a high-risk warning is triggered.

[0010] As a further solution of the present invention: The generation method of the visual warning report includes: Overlay the instruction evolution trajectory line with the historical health trajectory clusters, and use a dynamic color heat map to distinguish the risk levels; the report annotates the types of interaction features with the fastest degradation rates, including the frequency of logical breaks and the frequency of intention fluctuations, and correlates with the historical behavior records in the user's health file to compare the consistency between the current degradation trend and the previous medical diagnosis results; the visualization warning report includes key degradation time nodes. When a key degradation time node is clicked, the specific instruction data and execution results corresponding to that time node are traced back to assist in locating the source of abnormal behavior.

[0011] As a further solution of the present invention: all instruction data is locally completed for feature extraction and anonymization processing. The extraction process uses edge computing technology to ensure that the original instruction data does not leave the user terminal; the anonymized feature vectors are uploaded to the cloud database through an asymmetric encryption algorithm. The database adopts an independent partition storage architecture, with the user identity information physically isolated from the interaction data, and the access rights of each partition are controlled by an independent key; Medical parties need to access the complete report through double verification of the user authorization code and the system dynamic key. The validity period of the dynamic key is limited to a single session and automatically expires after timeout; after the data retention period ends, the instruction data stored in the cloud and locally will automatically trigger an irreversible destruction program.

[0012] As a further solution of the present invention: it also includes a verification mechanism: After generating a high-risk warning report, the system automatically extracts the instruction data of the user in multiple interaction scenarios within a set time period, including device control, complex task execution, and multi-round dialogue interaction; compare the logical coherence score distribution, intention rationality score decay curve, and response stability under different scenarios. If all scenarios show a consistent degradation trend, it is determined as the risk of disease progression; If only specific scenarios are abnormal, obtain environmental interference factors, including device response delay and network connection stability; if the environmental interference is the main cause, recalibrate the baseline model parameters and mark false alarm events.

[0013] The present invention also includes an intelligent Parkinson's disease dysfunction assessment system for implementing the above-mentioned intelligent Parkinson's disease dysfunction assessment method, including: A data collection module, based on deploying intelligent interaction terminals in the user's daily living environment, continuously collects unstructured user instruction data through the user's daily interaction behaviors with the devices; An instruction processing module, used to perform anonymization processing on the collected instruction data and extract a multi-dimensional feature set including instruction logical coherence and intention rationality; A model construction module that establishes a dynamic baseline model based on the instruction data of the user's health status period. The baseline model automatically expands the storage capacity according to the time dimension to form a personalized reference database that evolves synchronously with the user's interaction characteristics. A cognitive analysis module that analyzes the evolution matching degree between the real-time instruction features and the baseline model. When it detects that the deviation value of the multi-dimensional feature set continuously decreases and exceeds the adaptive threshold, it generates a visual warning report containing the disease progression risk level.

[0014] Advantages of the present invention: Based on the intelligent interaction terminal deployed in the user's daily living environment, the present invention comprehensively collects user instruction data such as voice and direct operations, solving the problem of limited data collection in the prior art; anonymizes the instruction data and extracts a multi-dimensional feature set containing logical coherence and intention rationality, which can analyze the user's functional status more accurately and objectively compared with traditional subjective judgments; establishes a dynamic baseline model based on the health status instruction data, which can naturally evolve with the user's normal behavior to achieve real-time dynamic evaluation; generates a visual warning report by calculating the evolution matching degree and setting the adaptive threshold, can timely detect the disease progression risk, make up for the defect that the existing methods cannot monitor the subtle progression of the disease in time, achieve early warning, and avoid the patient missing the best treatment opportunity. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The present invention will be further described below with reference to the accompanying drawings.

[0016] Figure 1 It is a schematic flowchart of an intelligent Parkinson's disease dysfunction assessment method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0018] Please refer to Figure 1 As shown, the present invention is an intelligent Parkinson's disease dysfunction assessment method, including the following steps: Data collection: In the user's daily living environment, an intelligent interaction terminal is carefully deployed. When the user operates various devices in daily use, the terminal automatically and continuously collects unstructured user instruction data. Whether it is a voice instruction issued by the user or a direct operation instruction, it can be accurately captured. Data processing and feature extraction: After collecting the instruction data, anonymization processing is immediately carried out on it. An algorithm is used to deeply extract a multi-dimensional feature set including the logical coherence and intention rationality of the instructions, and carefully distinguish the logical relationships and the rationality of the intentions in the instructions. Dynamic baseline model construction: Based on the instruction data collected during the user's healthy state, a dynamic baseline model is constructed. This baseline model has unique advantages. It can automatically expand the storage capacity according to the time dimension, closely fit the user's daily interaction habits, and gradually form a personalized reference database that evolves synchronously with the user's interaction characteristics. Evaluation and early warning: By deeply analyzing the evolution matching degree between the real-time instruction features and the baseline model, once it is detected that the logical coherence score continues to decline and the intention rationality score fluctuates abnormally exceeding the adaptive threshold, the system will quickly generate a visual early warning report including the risk level of disease progression, providing an intuitive and crucial reference basis for the evaluation of Parkinson's disease dysfunction.

[0019] In a preferred embodiment of the present invention, the collection of user instruction data: User instructions mainly cover two categories: voice instructions and direct operation instructions. The intelligent interaction terminal has a highly intelligent instruction collection activation mechanism. When the user engages in device control, such as smart home operations, easily adjusting the brightness of the lights and the temperature, or adjusting the parameters of electronic devices, such as adjusting the volume of the TV and the display resolution of the computer; when conducting information queries, actively asking about weather conditions, news, etc.; and when carrying out daily conversations, having voice conversations through the device, chatting about daily trivia and discussing hobbies, etc., the terminal will automatically and sensitively activate the instruction collection function, accurately intercepting the complete interaction instruction content of the user to ensure that no key information is missed. After collecting the instruction content, it will be carefully classified and marked according to the interaction scenario. For example, smart home operation instructions are marked as "home control category", information query instructions are marked as "information acquisition category", etc. These marked instructions will be stored in both the local encrypted cache area and the cloud database at the same time. The local encrypted cache area can store data in the first time, ensuring the immediacy and security of the data and preventing data loss before transmission. The cloud database only retains the instruction feature vectors after anonymization processing, and the feature vectors are desensitized through an advanced hash algorithm, further strengthening the defense line of data privacy protection and avoiding the risk of user information leakage.

[0020] In another preferred embodiment of the present invention, the process of extracting a multi-dimensional feature set including the logical coherence and intention rationality of the instructions: First, comprehensively collect the original data generated by the user through voice commands and direct operation commands. For voice commands, use efficient and accurate speech recognition technology for text conversion to accurately convert the words spoken by the user into text form. Regarding operation commands, record their types in detail, such as click operation, swipe operation, or long-press operation; accurately record the trigger time, accurate to the millisecond level for subsequent time series analysis; and completely record the action path, such as the trajectory of swiping from point A to point B on the screen. Immediately afterwards, strictly align the voice text and operation commands according to the millisecond-level timestamp to construct a multi-modal command sequence. In this sequence, it is possible to keenly identify multiple repetitions or contradictory expressions of the user's intention in the same task. For example, when setting an alarm, the user first says "set 7 am" by voice and then manually sets it to "8 am", and this contradictory expression can be accurately captured. Next, deeply detect the semantic contradictions between adjacent commands. Once found, immediately mark them as short-term logical conflicts. When there is an inconsistency in semantics or action direction between the voice command and the operation command, such as saying "turn on the TV" by voice but performing the operation of turning off the TV, it is marked as a cross-modal logical conflict. Then, count the frequency of logical conflict events per unit time. Different types of conflicts are assigned different weights according to their impact on the evaluation, and a logical coherence score is generated through a scientific calculation method to quantify the logical coherence degree between commands. In terms of intention rationality analysis, carefully analyze the context matching degree of the commands. Mark the commands issued without spatio-temporal trigger conditions, such as suddenly saying "travel tomorrow" without any time-related prompt, as spatio-temporal abnormal intentions; mark the commands with non-coherent operation chains, such as continuously clicking on different function modules illogically when operating a computer, as disordered abnormal intentions; when there is no coordination relationship between the voice command and the operation command within a time-intensive interval, such as asking for a recipe by voice while randomly swiping pictures on the mobile phone, it is determined as a chaotic abnormal intention. Similarly, different weights are assigned according to the types of abnormal intentions to generate an intention rationality score. Finally, the logical coherence score and the intention rationality score are skillfully fused into a multi-dimensional feature vector representing the user's cognitive degradation through a specific fusion algorithm, providing key data support for subsequent evaluation.

[0021] In another preferred embodiment of the present invention, a method for constructing a dynamic baseline model: 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.

[0022] First, these initial data must be strictly confirmed by professional medical staff to ensure that there are no interaction dysfunctions during data collection. Medical staff will comprehensively evaluate the user's physical condition and cognitive function by combining various medical examination methods and professional knowledge. Only after confirming that the user is in good health and can interact with the device normally will the collected data be incorporated into the initial model.

[0023] Secondly, the collection period should be extensive and comprehensive enough to cover the user's typical interaction behaviors at different time periods. For example, it is necessary to consider the different life rhythms and interaction habits of users on weekdays and weekends, as well as different activity patterns during the day and at night. This means that data collection may need to last for weeks or even months to ensure that various possible interaction scenarios and behavior patterns of users can be captured.

[0024] Over time, the user's interaction behaviors may undergo some natural changes. In order for the dynamic baseline model to accurately reflect the user's normal behaviors, regular updates are required. The update process mainly includes the following steps: Time series alignment: When new instruction data is added, these data need to be aligned with historical data in terms of time series first. The purpose of this step is to ensure that data from different time periods can be accurately corresponded in the time dimension for subsequent analysis, so as to conduct reasonable comparisons and calculations. For example, through information such as timestamps, the new data and historical data are arranged in chronological order to make the data at each time point accurately match.

[0025] Data segment division and deviation value calculation: The aligned data is divided using the sliding window technique, splitting it into data segments of a fixed length. Each data segment represents the user's interaction data within a specific time period. Then, for the data within each time window, the deviation value of the multi-dimensional feature set is calculated. This deviation value is calculated based on a comprehensive consideration of the logical coherence score and the intention rationality score. The logical coherence score reflects whether the logical relationship between user instructions is reasonable, while the intention rationality score measures whether the intention of the user's instructions conforms to common sense. By comprehensively considering these two scores, the characteristics of the user's interaction behavior can be evaluated more comprehensively.

[0026] Deviation value judgment and model update: A preset healthy range is set to determine whether the currently calculated deviation value is within the normal range. If the deviation value is within the preset healthy range, it indicates that there are no obvious abnormal changes in the user's interaction behavior. At this time, the new features can be added to the baseline model, and the logical coherence mean value and the fluctuation range are recalculated. This can ensure that the model can be continuously updated as the natural evolution of the user's normal behaviors, maintaining its adaptability to the user's current behavior pattern.

[0027] When the deviation value exceeds the healthy range, it indicates that there may be abnormal changes in the user's interaction behavior. At this time, the system will trigger an artificial review process. The content of the artificial review includes a detailed investigation of factors such as recent changes in the user's living environment, adjustments to device operation habits, or temporary mood swings. For example, the user may have changed their living environment due to moving, resulting in a change in the interaction method with smart home devices; or the user may have learned new device operation skills, thus changing their operation habits; or the user may have experienced mood swings due to high work pressure or other reasons recently, affecting the interaction performance with the device. Only after confirming that these non-disease factors are the causes of the deviation is it allowed to update the model to avoid the wrong influence of abnormal data caused by non-disease factors on the model.

[0028] In another preferred embodiment of the present invention, the calculation process of the evolution matching degree: Map the instruction features collected in real time into the interaction feature space constructed by the baseline model. This interaction feature space is a three-dimensional space, including a logical coherence scoring axis, an intention rationality scoring axis, and a response timeliness coordinate axis. By analyzing and calculating the real-time instruction features, the scores in the three aspects of logical coherence, intention rationality, and response timeliness are obtained respectively, and then these scores are used as coordinate values to generate the three-dimensional coordinate position of the current feature point in the interaction feature space. For example, assume that the logical coherence score of the real-time instruction is 80 points, the intention rationality score is 75 points, and the response timeliness score is 85 points. Then the three-dimensional coordinates of this feature point are (80, 75, 85).

[0029] Based on the time dimension, connect the feature points in consecutive acquisition cycles in sequence to form an instruction evolution trajectory line. This trajectory line can intuitively reflect the trend of the user's interaction features changing over time. For example, if the logical coherence score of the user gradually decreases over a period of time, the trajectory line will show a downward trend on the logical coherence scoring axis; if the intention rationality score fluctuates greatly, the trajectory line will show a more tortuous shape on the intention rationality scoring axis. By observing the shape and trend of the trajectory line, the change situation of the user's interaction behavior can be initially judged.

[0030] To more accurately evaluate the degree of difference between the current trajectory line and the interaction behavior in the user's historical healthy state, it is necessary to calculate the minimum enclosing distance between the trajectory line and the historical healthy trajectory cluster. The historical healthy trajectory cluster is composed of multiple typical trajectories of the user in the healthy state, and 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 is, the closer the current trajectory line is to the historical healthy trajectory, and the more normal the user's interaction behavior is; on the contrary, the larger the distance is, the more likely it indicates that the user's interaction behavior has undergone abnormal changes.

[0031] When the trajectory line continuously deviates from the historical healthy trajectory cluster, and the deviation direction shows a logical break in the instruction accompanied by an increase in the response delay rate, it is determined as an abnormal degradation mode. The logical break can be quantified by the jump frequency of the instruction context, that is, by counting the number of jumps or incoherences in the context relationship between instructions within a certain time period. The response delay rate is calculated by the time difference of the device executing instructions, that is, by recording the time interval between the user issuing an instruction and the device actually executing the instruction, and calculating the average delay time. If the logical break frequency is high and the response delay rate also increases significantly, it indicates that the user's interaction ability may have degraded, and there is a risk of Parkinson's disease dysfunction.

[0032] In another preferred embodiment of the present invention, the method for setting the adaptive threshold: According to the historical degradation law of the logical coherence score and the fluctuation trend of the intention rationality score in the baseline model, a dynamic threshold curve is established. The shape of this curve is jointly determined by the decline rate of the logical coherence score and the abnormal fluctuation amplitude of the intention rationality score. Specifically, through the analysis of historical data, the decline rate of the logical coherence score and the abnormal fluctuation amplitude of the intention rationality score in different time periods are statistically analyzed, and then a curve is fitted based on these statistical data. For example, if in some time periods, the logical coherence score drops rapidly and the intention rationality score fluctuates abnormally greatly, then on the curve corresponding to these time periods, the threshold will be increased accordingly to adapt to this change.

[0033] To more accurately reflect the importance of the logical coherence score and the intention rationality score in risk assessment, corresponding weight coefficients need to be set for these two scores. The determination of the weight coefficients can be carried out according to medical research and practical experience. For example, if through a large number of experiments and analyses, it is found that the logical coherence score is more critical in the assessment of Parkinson's disease dysfunction, then a higher weight coefficient can be set for the logical coherence score.

[0034] When the decline rate of the logical coherence score exceeds the current threshold curve, it indicates that there may be abnormal changes in the user's interaction behavior. At this time, a fuzzy logic algorithm is used to calculate the comprehensive risk index. The fuzzy logic algorithm is an algorithm that can process uncertain and fuzzy information. Its input variables include the logical break strength, the abnormal fluctuation frequency of the intention rationality, and the cross-modal conflict frequency. By performing operations such as fuzzification, rule inference, and defuzzification on these input variables, a risk level is output and mapped to the interval [0, 1]. For example, a risk level of 0 means there is no risk at all, and a risk level of 1 means the risk is extremely high. A critical value is set. When the calculated comprehensive risk index value breaks through this critical value, the system will trigger a high-risk warning. This means that the system believes that the user has a relatively high risk of Parkinson's disease functional impairment progression and further examinations and intervention measures need to be taken in a timely manner.

[0035] In another preferred embodiment of the present invention, a method for generating a visual warning report: Precisely superimpose and compare the instruction evolution trajectory line with the historical health trajectory cluster, making the changes in the user's interaction behavior clear at a glance. To more clearly distinguish the risk levels, the system adopts a dynamic color heat map technology, and different risk levels will be visually presented in different colors. For example, the low-risk area is shown in green, the medium-risk area is shown in yellow, and the high-risk area is shown in red, enabling the user to immediately identify the current risk level. The report will also detail the type of interaction feature with the fastest degradation rate, which may be the frequency of logical breaks, that is, the number of times the instruction logic breaks within a unit time; the response delay duration, accurate to the millisecond level of the device response delay time; or the execution error rate, that is, the proportion of the number of incorrect instruction executions in the total number of instruction executions. At the same time, the report will also intelligently associate with the historical behavior records in the user's health file, carefully compare the current degradation trend with the previous medical diagnosis results to see if they are consistent. The report specifically includes key degradation time nodes. When the user clicks on these key degradation time nodes, the system will quickly trace back the specific instruction content and execution results during that period, which can accurately assist in locating the source of abnormal behavior and provide strong data support for subsequent in-depth analysis of the condition and formulation of intervention measures.

[0036] In another preferred embodiment of the present invention, all instruction data, from the very beginning of collection, undergoes a comprehensive and efficient feature extraction and anonymization process locally. In the feature extraction stage, cutting-edge edge computing technology is adopted, leveraging the computing power of the user terminal device itself to directly conduct in-depth mining on the original instruction data locally, extracting features including key information such as the logical coherence of instructions and the rationality of intentions. This local processing method maximally ensures that the original instruction data does not leave the user terminal, eliminating the potential risk of data leakage during transmission from the source. After feature extraction is completed, anonymization processing is immediately carried out on the data. Through complex and reliable algorithms, all sensitive information that can identify the user's identity is removed, leaving only the necessary features for Parkinson's disease dysfunction assessment. The anonymized feature vectors are uploaded to the cloud database using an asymmetric encryption algorithm. The asymmetric encryption algorithm is like an indestructible "digital armor" for the data. Different keys are used for the encryption and decryption processes, greatly enhancing the security of data transmission. The cloud database adopts a carefully designed independent partition storage architecture, and the interaction data of each user is stored separately in an independent partition. More importantly, the user identity information and the interaction data are physically isolated, just like placing the two in different "safe boxes", further ensuring data security. Moreover, the access rights of each partition are strictly controlled by independent keys, and only authorized personnel holding the corresponding correct keys can access the data in a specific partition. This refined permission management mode effectively prevents the illegal access and abuse of data. Regarding data access rights, if the medical party wants to obtain a complete report, it needs to go through a strict dual-verification process of the user authorization code and the system dynamic key. The user authorization code is a permission credential actively granted by the user based on trust in the medical party, ensuring the legality and user autonomy of data access. The system dynamic key generates a unique temporary key for each access, and its validity period is strictly limited to a single session. Once it times out, the dynamic key will automatically become invalid. Just like a "digital door lock" with a time limit, even if the key information is accidentally leaked, it cannot be used to access the data after timeout, adding another solid defense line for data security. When the data retention period ends, whether it is the instruction data stored in the cloud or the relevant data cached locally, an irreversible destruction program will be automatically triggered. This program uses complex algorithms to overwrite and erase the data multiple times to ensure that the data cannot be recovered, completely eliminating the potential risks brought by data retention. In another preferred embodiment of the present invention, the verification mechanism includes: After generating a high-risk warning report, a thorough verification process will be automatically initiated immediately. The system will comprehensively extract the instruction data generated by the user under various interaction scenarios within a set time period. These interaction scenarios cover device control, such as when the user operates smart home devices for turning on / off, adjusting, etc.; complex task execution, like 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-round dialogue interaction, for example, having continuous multi-round question-and-answer exchanges with a smart voice assistant. After extracting the data, the system will carefully compare the logical coherence of instructions in different scenarios, judge whether the semantic and logical associations between instructions are tight; execution accuracy, by counting the proportion of correctly executed instructions; and response stability, by checking whether the device responds to instructions stably and smoothly, without delays or abnormal interruptions. If there is a consistent degradation trend in all these different scenarios, such as worse instruction logical coherence, lower execution accuracy, and decreased response stability, etc., the system will cautiously determine it as a risk of disease progression, providing a strong basis for medical decision-making. However, if abnormal conditions occur only in specific scenarios, the system will not jump to conclusions. At this time, the system will quickly obtain possible environmental interference factors, including device response delay, that is, the time taken for the device to start executing after receiving the instruction is too long, which may be due to the device's own performance issues or handling too many tasks simultaneously; network connection stability, if the network signal is poor, frequently interrupted, or has too high a delay, it may also affect the user's interaction with the device. If it is determined through analysis that the environmental interference is the main cause, the system will promptly recalibrate the baseline model parameters, optimize and adjust the model according to the current actual environmental factors to make it more in line with the actual situation. At the same time, the system will rigorously mark false alarm events and record the detailed information of this abnormal situation for subsequent analysis and summary, continuously improving the accuracy and reliability of the evaluation system.

[0037] The present invention also includes an intelligent Parkinson's disease dysfunction evaluation system for implementing the above-mentioned intelligent Parkinson's disease dysfunction evaluation method, including: A data acquisition module, which is based on deploying intelligent interaction terminals in the user's daily living environment and continuously acquires unstructured user instruction data through the user's daily interaction behaviors with the devices; An instruction processing module for anonymizing the acquired instruction data and extracting a multi-dimensional feature set including instruction logical coherence and intention rationality; A model construction module that establishes a dynamic baseline model based on the instruction data during the user's healthy state, and the baseline model automatically expands the storage capacity according to the time dimension to form a personalized reference database that evolves synchronously with the user's interaction characteristics; A cognitive analysis module, which is used to generate a visual warning report containing the risk level of disease progression by analyzing the evolution matching degree between real-time instruction features and a baseline model when it detects that the deviation value of a multi-dimensional feature set continuously decreases and exceeds an adaptive threshold.

[0038] The above has described an embodiment of the present invention in detail, but the content described is only a preferred embodiment of the present invention and cannot be considered as limiting the scope of implementation of the present invention. All equal changes and improvements made in accordance with the scope of the present invention application should still fall within the scope covered by the patent of the present invention.

Claims

1. An intelligent method for assessing Parkinson's disease functional impairment, characterized in that: The following steps are involved: Deploy smart interactive terminals in users' daily living environments to continuously collect unstructured user command data through daily interactions between users and devices; Anonymize the collected instruction data and extract a multi-dimensional feature set that includes the logical coherence and rationality of the instruction intention; Establishing a dynamic baseline model based on the instruction data of the user's health status period, the baseline model automatically expands the storage capacity according to the time dimension to form a personalized reference database that evolves synchronously with the user's interaction characteristics; By analyzing the evolutionary match between real-time instruction features and the baseline model, when it is detected that the deviation value of the multidimensional feature set continues to decrease and exceeds the adaptive threshold, a visual early warning report containing the risk level of disease progression is generated.

2. An intelligent Parkinson's disease functional impairment assessment method according to claim 1, characterized in that: The collection of the user instruction data specifically includes: The user instructions include voice instructions and direct operation instructions. When the user performs device control, information query or daily communication, the intelligent interactive terminal automatically activates the instruction collection function and intercepts the complete interactive instruction content of the user; the device control includes smart home operation and electronic equipment parameter adjustment, the information query is the user's active query instruction, and the daily communication is the device voice question and answer; The command content is classified and labeled according to the interaction scenario and stored in the local encrypted cache area and cloud database; the cloud database only retains the anonymized command feature vector, and the feature vector is desensitized through a hash algorithm.

3. An intelligent Parkinson's disease functional impairment assessment method according to claim 2, characterized in that: The process of extracting a multidimensional feature set that includes the logical coherence of instructions and the rationality of intentions is: Collect the raw data generated by users through voice commands and direct operation commands, convert voice commands into text, and record the type, trigger time and action path of operation commands; align voice text and operation commands by millisecond timestamps, construct multimodal command sequences, and identify multiple repetitions or contradictory expressions of user intentions in the same task; Detect semantic contradictions between adjacent instructions and mark them as short-term logic conflicts; When the voice instructions and the operation instructions are inconsistent in semantics or action direction, it is marked as a cross-modal logic conflict; Count the frequency of logical conflict events per unit time, assign difference weights according to the conflict types, and generate a logical coherence score; Analyze the context matching of the instructions, mark the instructions issued without spatiotemporal trigger conditions as spatiotemporal abnormal intentions, mark the instructions of incoherent operation chains as disordered abnormal intentions, and judge them as disordered abnormal intentions when the voice instructions and operation instructions have no synergistic relationship in the time-intensive interval. Assign difference weights according to the types of abnormal intentions to generate intention rationality scores; The logical coherence score and intention rationality score are fused into a multidimensional feature vector to characterize user cognitive degradation.

4. The intelligent Parkinson's disease functional impairment assessment method according to claim 3, characterized in that: The method for constructing the dynamic baseline model includes: The initial model uses the health status instruction data collected continuously when the user first registers. The initial data has been confirmed by medical staff to have no interactive function impairment, and the collection cycle covers the user's typical interactive behaviors in different time periods. When the model is updated, the newly added instruction data is aligned with the historical data in time series, the data segments are divided by sliding window technology, and the deviation value of the multidimensional feature set in each time window is calculated; the deviation value is calculated based on the logical coherence score and the intention rationality score; when the deviation value is in the preset healthy range, the new feature is added to the baseline model and the logical coherence mean and fluctuation range are recalculated to ensure that the model evolves naturally with the normal behavior of the user; When the deviation value exceeds the healthy range, the manual review process is triggered. The review content includes the user's recent changes in living environment, adjustments in equipment operation habits, or temporary mood swings. The model is only allowed to be updated after confirming that the deviation is not caused by disease factors.

5. The intelligent Parkinson's disease functional impairment assessment method according to claim 1, characterized in that: The calculation process of the evolution matching degree includes: Mapping the real-time instruction features to the interactive feature space constructed by the baseline model, wherein the feature space includes a logical coherence scoring axis, an intention rationality scoring axis, and a response timeliness coordinate axis, and generating a three-dimensional coordinate position of the current feature point; Connecting feature points of consecutive acquisition cycles based on the time dimension to form a command evolution trajectory, wherein the trajectory reflects the trend of user interaction features changing over time; Calculating the minimum enclosing distance between the trajectory line and a historical health trajectory cluster, where the historical health trajectory cluster is composed of multiple typical trajectories under the user's health state, representing the fluctuation range of normal interactive behavior; When the trajectory line continues to deviate and the deviation direction is manifested as a command logic break accompanied by an increase in the response delay rate, it is determined to be an abnormal degradation mode; the logic break is quantified by the command context jump frequency, and the response delay rate is calculated by the time difference of the device executing the command.

6. The intelligent Parkinson's disease functional impairment assessment method according to claim 1, characterized in that: The method for setting the adaptive threshold comprises: According to the historical degradation law of the logical coherence score and the fluctuation trend of the intention rationality score in the baseline model, a dynamic threshold curve is established, and the shape of the curve is jointly determined by the decline rate of the logical coherence score and the abnormal fluctuation amplitude of the intention rationality score; corresponding weight coefficients are set for the logical coherence score and the intention rationality score; When the rate of decrease of the logical coherence score exceeds the current threshold curve, the fuzzy logic algorithm is used to calculate the comprehensive risk index. The input variables of the algorithm include the strength of logical fracture, the frequency of abnormal fluctuations in intention rationality, and the frequency of cross-modal conflicts. 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.

7. The intelligent Parkinson's disease functional impairment assessment method according to claim 1, characterized in that: The method for generating the visual early warning report includes: The command evolution trajectory line is superimposed and compared with the historical health trajectory cluster, and a dynamic color heat map is used to distinguish the risk level; the report marks the interactive feature type with the fastest degradation rate, including the frequency of logical breaks and the frequency of intention fluctuations, and links the historical behavior records in the user's health file to compare the consistency of the current degradation trend with the previous medical diagnosis results; the visual early warning report includes key degradation time nodes. When the key degradation time node is clicked, the specific command data and execution results of the corresponding time node are traced back to assist in locating the source of abnormal behavior.

8. The intelligent Parkinson's disease functional impairment assessment method according to claim 1, characterized in that: All instruction data are locally feature extracted and anonymized. The extraction process uses edge computing technology to ensure that the original instruction data does not leave the user terminal. The anonymized feature vector is uploaded to the cloud database through an asymmetric encryption algorithm. The database adopts an independent partition storage architecture. The user identity information is physically isolated from the interactive data, and the access rights of each partition are controlled by an independent key. The medical party needs to double-verify the user authorization code and the system dynamic key to access the complete report. The dynamic key is valid for a single session and will automatically expire after timeout. After the data retention period ends, the command data stored in the cloud and locally will automatically trigger an irreversible destruction procedure.

9. The intelligent Parkinson's disease functional impairment assessment method according to claim 1, characterized in that: Also includes authentication mechanisms: After generating a high-risk warning report, the system automatically extracts the user's command data in various interaction scenarios within a set time period, including device control, complex task execution, and multi-round dialogue interaction; compares the logical coherence score distribution, intention rationality score attenuation curve, and response stability in different scenarios. If all scenarios show a consistent degradation trend, it is determined to be a risk of disease progression. If the anomaly occurs only in a specific scenario, obtain environmental interference factors, including device response delay and network connection stability; If environmental interference is the main cause, the baseline model parameters are recalibrated and false positive events are marked.

10. An intelligent Parkinson's disease functional impairment assessment system, used to implement an intelligent Parkinson's disease functional impairment assessment method according to any one of claims 1 to 9, characterized in that: include: The data collection module deploys intelligent interactive terminals in the user's daily living environment and continuously collects unstructured user command data through the daily interaction between the user and the device; The instruction processing module is used to anonymize the collected instruction data and extract a multi-dimensional feature set including the logical coherence and rationality of the instruction intention; A model building module, which builds a dynamic baseline model based on the instruction data of the user's health status period, and the baseline model automatically expands the storage capacity according to the time dimension to form a personalized reference database that evolves synchronously with the user's interaction characteristics; The cognitive analysis module is used to analyze the evolutionary matching between the real-time instruction features and the baseline model, and generate a visual early warning report containing the risk level of disease progression when it is detected that the deviation value of the multidimensional feature set continues to decrease and exceeds the adaptive threshold.

Images