Mild cognitive impairment detection system based on multilevel data

Abstract

The invention relates to the technical field of data processing, and discloses a mild cognitive impairment detection system based on multi-level data, and the system comprises a terminal device which is used for collecting physiological monitoring data, clinical narrative data and multi-modal test data of a user; the data processing module is used for performing interference identification and information purification on the multi-modal test data to generate a purified cognitive response matrix; the cognitive evaluation module is used for performing cognitive ability decoupling and group benchmarking based on the purified cognitive response matrix and outputting an individual ability level; and the closed-loop management module is used for constructing a longitudinal decline slope model based on a score sequence of individual historical multiple evaluations and carrying out risk early warning and intervention recommendation. According to the method, passive disease screening of cognitive health management is converted into an active, personalized and continuous risk monitoring and intervention process, and a powerful technical tool is provided for delaying cognitive decline.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a mild cognitive impairment detection system based on multi-level data. Background Technology

[0002] Cognitive impairment, particularly mild cognitive impairment (MCI) caused by neurodegenerative diseases such as Alzheimer's, has become a major global public health challenge. Early identification and intervention are crucial to slowing disease progression. However, existing cognitive assessment methods have many limitations. Traditional neuropsychological scales (such as MMSE and MoCA) are typically administered by physicians in clinical settings, which is time-consuming and the results are easily affected by the subjective factors of the test taker, the patient's condition at the time (such as fatigue and mood), and the decline in sensory organ functions (such as vision and hearing), leading to decreased reliability and validity. These assessments are mostly "snapshot-like," lacking continuous tracking of the longitudinal trend of individual cognitive abilities, making it difficult to capture the slow decline trajectory in the early stages. Furthermore, there is often a disconnect between assessment results and personalized, actionable intervention recommendations, failing to form an effective closed loop for health management.

[0003] In recent years, with the development of wearable devices and mobile computing technologies, some cognitive assessment schemes based on digital biomarkers have emerged. However, these schemes typically rely on a single data source (such as motion data from a wristband or touchscreen interaction from a mobile phone), failing to fully utilize the rich information provided by multimodal data (such as eye movements, speech, and gestures). Moreover, existing technologies cannot eliminate confounding factors (such as drug side effects and sleep deprivation) that can pollute cognitive test signals, resulting in significant biases in the assessment results.

[0004] Therefore, there is an urgent need for a personal cognitive impairment detection and risk assessment system that can achieve high-precision, personalized, continuous dynamic assessment and form intervention management. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides a mild cognitive impairment detection system based on multi-level data, which at least partially solves the above-mentioned technical problems.

[0006] This invention provides a mild cognitive impairment detection system based on multi-level data, comprising: Terminal devices used to collect user physiological monitoring data, clinical narrative data, and multimodal test data; A data processing module for performing interference identification and information purification on the multimodal test data to generate a purified cognitive response matrix; A cognitive assessment module for decoupling cognitive abilities and benchmarking against groups based on the purified cognitive response matrix, and outputting individual ability levels. This is a closed-loop management module used to construct a longitudinal decline slope model based on a scoring sequence of multiple historical assessments of an individual, and to provide risk warnings and intervention recommendations.

[0007] Compared with the prior art, the present invention has the following technical effects: The technical advantages of this invention are as follows: For the first time, it integrates the collection of multi-source heterogeneous data, intelligent interference filtering based on medical knowledge, decoupled cognitive ability assessment, individualized trend prediction based on time-series data, and intervention recommendations into a single automated process. This effectively overcomes the limitations of traditional single-time, single-dimensional assessments, which are susceptible to transient state interference, fail to reflect changing trends, and are disconnected from intervention. Through the synergy of hardware and algorithms, the system can achieve near-clinical-level assessment reliability in community or home settings, and significantly improves the early identification sensitivity of mild cognitive impairment by capturing microscopic signs of decline. Ultimately, it transforms cognitive health management from passive disease screening to a proactive, personalized, and continuous risk monitoring and intervention process, providing a powerful technical tool for delaying cognitive decline.

[0008] Furthermore, by introducing a rule-based interference identification mechanism, it is possible to proactively and in real-time identify and label test performance biases caused by the user's own physiological or pathological state (rather than cognitive ability), enhancing the system's adaptability and fairness. This ensures that the assessment of cognitive abilities of individuals with different comorbidities or taking different medications can be conducted on a relatively "pure" and comparable basis. This effectively solves the "signal contamination" problem commonly found in traditional digital assessments.

[0009] Furthermore, the question selection process removes data points with a high risk of systematic bias; the repeatability and consistency checks filter out random noise interference; and the cognitive response matrix can strip away various known and unknown non-cognitive confounding factors, fusing weak, real cognitive ability signals scattered in multimodal signals. This matrix directly determines the accuracy, stability, and repeatability of the entire system evaluation results. Attached Figure Description

[0010] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of the structure of a mild cognitive impairment detection system based on multi-level data provided in an embodiment of the present invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0013] This invention provides a mild cognitive impairment detection system based on multi-level data, including a terminal device with communication connection, a data processing module, a cognitive assessment module, and a closed-loop management module. The function of each module is described in detail below.

[0014] Terminal devices include smart bracelet terminals, mobile phone terminals, and multimodal testing equipment. Smart bracelet terminals are used to collect users' physiological monitoring data, including sleep stage data, motor behavior data, and autonomic nervous system data; mobile phone terminals are used to acquire clinical narrative data, including medication records, historical diagnostic results, and travel trajectories; multimodal testing equipment includes an embedded industrial computer, microphone array, eye tracker, touchscreen, and camera, used to perform multimodal cognitive tests and collect multimodal test data.

[0015] Specifically, the smart bracelet terminal, acting as a wearable sensing node, integrates a high-precision accelerometer, a PPG optical heart rate sensor, a gyroscope, and an ambient light sensor. Through sensor fusion algorithms, it continuously and seamlessly collects the user's physiological monitoring data. For example, the smart bracelet terminal analyzes nighttime body movement and heart rate variability to generate sleep stage data, including the percentage of deep sleep, REM sleep latency, and the number of micro-awakenings; it analyzes daytime acceleration signals to calculate motor behavior data, such as average daily steps, gait regularity, and body balance indicators; simultaneously, it continuously monitors the time-frequency characteristics of heart rate variability (HRV) and skin conductance levels as autonomic nervous system data. This data is periodically and encryptedly transmitted to the user's mobile terminal via Bluetooth Low Energy protocol.

[0016] After obtaining explicit user authorization, a dedicated application installed on the mobile terminal automatically retrieves structured clinical narrative data by calling the operating system's health suite interface or a secure data exchange interface with the healthcare information system (HIS). This clinical narrative data includes a detailed list of medication records, specifically: drug name, dosage, administration time, drug half-life, and central nervous system penetration index; it also includes historical diagnostic results. The application uses natural language processing technology to parse structured diagnostic codes (such as ICD-10) and key comorbidity descriptions from the electronic health records. In the background, the dedicated application uses location data to determine the user's travel trajectory.

[0017] Multimodal testing equipment is a fixed integrated terminal. Its core hardware is an embedded industrial control computer based on an ARM architecture, responsible for overall control and real-time calculation. The equipment integrates a high-fidelity microphone array for clear sound pickup; a high-sampling-rate (e.g., 120Hz) infrared eye tracker for precise eye movement coordinate tracking; a high-sensitivity multi-touch LCD screen for displaying test questions and capturing fine clicks, swipes, and other gestures; and a wide-angle high-definition camera for recording facial expressions and upper body posture during the test. With brief guidance from professionals or clear self-guided prompts, users complete a series of standardized cognitive tasks on this device. The device simultaneously collects speech waveforms, eye movement sequences, touch event streams, and video streams, collectively forming the raw multimodal test data. For example, if the multimodal testing equipment displays test questions on the touchscreen, requiring the user to speak and follow the movement with their eyes, select sentences, and simulate actions, the device can simultaneously collect speech waveforms, eye movement sequences, touch event streams, and video streams.

[0018] The system collects multi-level user data through integrated terminal devices, including: physiological monitoring data generated by continuous monitoring from smart bracelets; clinical narrative data obtained by mobile applications through authorized interfaces; and multi-modal testing data, such as speech, eye movement, gesture, and touch sequences, collected by specialized multi-modal testing equipment performing standardized cognitive tasks in a controlled environment. This heterogeneous data is synchronously transmitted to the system's data processing module.

[0019] The data processing module performs interference identification and information purification on the multimodal test data to generate a purified cognitive response matrix.

[0020] Optionally, interference identification is performed on the multimodal test data, including: interference identification is performed through a preset causal rule base, which defines the mapping relationship between interference keywords and data operation instructions, and is used to match interference keywords with user physiological monitoring data and clinical narrative data; when the match is successful, the data operation instructions corresponding to the interference keywords are executed to freeze or weighted correct specific fields in the multimodal test data.

[0021] The causal rule base associates specific interference keywords (such as specific diseases or drugs) with operation instructions (such as freezing or weighting) on ​​multimodal test data fields. The data processing module matches the user's physiological monitoring data and clinical narrative data with interference keywords in the causal rule base in real time. Once a known interference source is identified, it performs preset operations on the corresponding multimodal test data fields.

[0022] In one specific implementation, the causal rule base consists of a series of rule entries. Each rule is a triple, including: 1) interference keywords (derived from clinical narrative data and physiological monitoring data), such as ICD codes for medical history diagnoses (e.g., "glaucoma"), drug names (e.g., "benzodiazepine"), or threshold conditions from wristband data (e.g., "deep sleep percentage" or "deep sleep <1.5h"); 2) data manipulation instructions, specifically freezing or weighted correction; freezing a data field completely excludes its use in subsequent analysis, typically used when the field has been determined to be completely destroyed by some interference; weighted correction applies a weighted regression factor to the data field to offset the negative effects of the drug. This is used when interference partially affects signal quality, and the weights can be empirical values; 3) the confidence weight of the rule (predefined), used for priority ranking or weighted fusion when multiple rules are triggered.

[0023] When the system performs cognitive tests, the data processing module synchronizes physiological monitoring data and clinical narrative data in real time and traverses the causal rule base to check for matching rule entries. According to the causal rule base, when a user's medical history includes "glaucoma," the system freezes all indicators in the eye movement data that rely on high visual contrast for effective evaluation (such as certain fine fixation point recognition), retaining only less affected indicators, such as spatial saccade amplitude and fixation stability. This avoids misdiagnosing eye movement abnormalities caused by visual impairment as cognitive impairment. According to the causal rule base, if the medication list includes sedative drugs with a half-life greater than 24 hours and central penetration greater than 0.6 (such as certain benzodiazepine derivatives), the test data for multiple modalities are multiplied by a weighted regression factor. Specifically, the weighted regression factor (e.g., 0.6) is multiplied by the fundamental frequency jitter amplitude in the speech modality; the weighted regression factor (e.g., 0.6) is multiplied by the delay time in the gesture modality. In subsequent calculations, these affected original multimodal test data will be replaced with weighted corrected values.

[0024] Next, the data processing module refines the multimodal test data to generate a refined cognitive response matrix. Optionally, it determines the sensitivity score of each test item to interference from the interference source and removes test data of test items susceptible to interference from the current interference source based on the sensitivity score; it performs multiple repeated tests on test items of the same cognitive dimension after form fine-tuning and calculates the response consistency coefficient; if the response consistency coefficient is lower than a set value, the test data of the corresponding test item is removed; it dynamically generates fusion weights for each modality of test data through an individualized interference suppression network; and it performs weighted fusion of each modality of test data based on the fusion weights to obtain the refined cognitive response matrix.

[0025] This embodiment, after freezing or weighting certain fields in the multimodal test data, continues to clean the data using an information purification algorithm. Interference sources refer to factors that may affect cognitive test results but are not themselves a reflection of cognitive ability. Examples include hearing impairment, vision loss, coordination problems, sleep deprivation, use of certain medications (such as sedatives), or emotional fluctuations. These factors can contaminate the test data, leading to misjudgments of true cognitive ability. Sensitivity score: This measures the degree to which the performance on a particular test item is affected by a specific interference source. The higher the score, the more easily the result of that item (i.e., the test data) is interfered with by the corresponding non-cognitive factors. Each test item has a pre-set sensitivity score for various interference sources. The current interference source can be collected from user physiological monitoring data and clinical narrative data. The current interference source is matched against the interference sources of each test item to obtain the sensitivity score for each test item affected by the current interference source. If the sensitivity score is greater than a set threshold, it indicates susceptibility to the current interference source, and the test data for that test item is removed (or frozen). Fine-tuning of the test questions includes: different sequences of numbers remembered, different colors of presented graphics, etc. After the user completes multiple test questions before and after the fine-tuning, the system calculates the consistency coefficient of the user's performance on these questions, such as calculating the coefficient of variation of three responses or the consistency of test data. If the calculated response consistency coefficient is lower than a set value (e.g., 0.7), it indicates that the user's response to this set of questions is unstable and may be affected by transient, random factors. Therefore, the data for this entire set of questions will be considered unreliable and discarded.

[0026] The test data validated in the first two stages proceeds to the final information fusion step. The system invokes a personalized interference suppression network to process the previously processed test data. This personalized interference suppression network is a lightweight neural network model. It takes user physiological monitoring data, clinical narrative data, and the modality of the test items as input and dynamically outputs a set of weights. These weights determine how the contribution of each modality's test data should be adjusted when fusing multimodal data (such as voice, eye movement, and touch) to assess cognitive ability, in order to suppress the specific interference the user is currently experiencing.

[0027] Specifically, the inputs to this personalized interference suppression network include: 1) feature embeddings of user physiological monitoring data (such as sleep quality indicators from the wristband); 2) feature embeddings of clinical narrative data (such as vectors generated from medication records); and 3) feature embeddings of the modality to which the test items belong (modalities include: voice, eye movement, gesture, touch, and correctness of answers). Based on these inputs, the personalized interference suppression network dynamically generates optimal fusion weights for the current user and the modality to which the current test item belongs. For example, for a user whose wristband data shows poor sleep, the personalized interference suppression network automatically reduces the weight of test data belonging to the gesture modality. Finally, the system uses this set of dynamic weights to perform a weighted summation of the (normalized) multimodal test data for all test items, generating a comprehensive, refined feature vector, i.e., a refined cognitive response matrix.

[0028] The cognitive assessment module performs cognitive ability decoupling and group benchmarking based on the refined cognitive response matrix, outputting individual ability levels. The cognitive assessment module includes a multi-head cognitive decoupling module, which processes the refined cognitive response matrix through a neural network and outputs scores for multiple individual abilities; these abilities include: working memory, attention, language ability, executive function, visuospatial ability, and orientation.

[0029] The multi-head cognitive decoupling module includes multiple parallel cognitive structure sub-modules (i.e., sub-neural network models, including convolutional layers and fully connected layers). The convolutional layers extract features representing different individual abilities from the purified cognitive response matrix, and the fully connected layers obtain scores for different individual abilities.

[0030] After outputting scores for multiple individual abilities, the system compares each ability score with reference distributions for groups of the same age, education level, and gender to obtain a score reflecting the individual's ability distribution level. Through comparison, the system outputs easily understandable standardized scores, clearly indicating the user's position relative to healthy peers in each cognitive dimension.

[0031] Finally, the closed-loop management module constructs a longitudinal decline slope model based on the scoring sequence of multiple historical assessments of an individual, and provides risk warnings and intervention recommendations.

[0032] Specifically, the closed-loop management module continuously collects standardized scores from multiple historical user assessments to form a scoring sequence (i.e., a time series).

[0033] Optionally, the longitudinal decline slope model is a linear mixed-effects model. The linear mixed-effects model estimates the fixed-effects slope and random-effects slope of an individual's cognitive ability by fitting a series of ratings from multiple historical assessments. The closed-loop management module calculates an individualized slope value based on the fixed-effects slope and random-effects slope. If the individualized slope value is greater than a set threshold, a decline risk warning is issued.

[0034] For example, for users The scoring sequence for this assessment is as follows: ; in, This is the score of the Tth evaluation.

[0035] The linear mixed-effects model (LME) is as follows: ; in, and These are fixed-effects parameters, representing the mean intercept and mean slope (i.e., fixed-effects slope) of the entire rating sequence. They are obtained through aggregation estimation based on all sample data, for example, using restricted maximum likelihood estimation. It is the score of the i-th evaluation. These are the random effects parameters, representing the random intercept and random slope (i.e., the random effects slope) of the i-th evaluation, respectively. It is the residual term. It is the time point of the i-th evaluation.

[0036] The individualized slope value is: ; like If the condition is determined to be accelerating, a warning of the risk of decline needs to be issued to users.

[0037] in, It decreases by 0.08 standard deviations per year.

[0038] Optionally, the closed-loop management module also includes an intervention recommendation engine, used to determine and recommend interventions to users based on individualized slope values, user physiological monitoring data, and clinical narrative data. For example, it can extract suitable interventions from a pre-set database, including cognitive training programs or lifestyle modification guidance, and push them to users via a mobile application.

[0039] The technical advantages of this invention lie in its first-ever integration of multi-source heterogeneous data collection, intelligent interference filtering based on medical knowledge, decoupled cognitive ability assessment, individualized trend prediction based on time-series data, and intervention recommendations into a single automated process. This effectively overcomes the limitations of traditional single-time, single-dimensional assessments, which are susceptible to transient state interference, fail to reflect changing trends, and are disconnected from intervention. Through the synergy of hardware and algorithms, the system can achieve near-clinical-level assessment reliability in community or home settings, and significantly improves the early identification sensitivity of mild cognitive impairment by capturing microscopic signs of decline. Ultimately, it transforms cognitive health management from passive disease screening to a proactive, personalized, and continuous risk monitoring and intervention process, providing a powerful technical tool for delaying cognitive decline.

[0040] Furthermore, by introducing a rule-based interference identification mechanism, it is possible to proactively and in real-time identify and label test performance biases caused by the user's own physiological or pathological state (rather than cognitive ability), enhancing the system's adaptability and fairness. This ensures that the assessment of cognitive abilities of individuals with different comorbidities or taking different medications can be conducted on a relatively "pure" and comparable basis. This effectively solves the "signal contamination" problem commonly found in traditional digital assessments.

[0041] Furthermore, the question selection process removes data points with a high risk of systematic bias; the repeatability and consistency checks filter out random noise interference; and the cognitive response matrix can strip away various known and unknown non-cognitive confounding factors, fusing weak, real cognitive ability signals scattered in multimodal signals. This matrix directly determines the accuracy, stability, and repeatability of the entire system evaluation results.

[0042] It should be noted that the terminology used in this invention is for describing specific embodiments only and is not intended to limit the scope of this application. As shown in this specification, unless the context clearly indicates otherwise, words such as "a," "an," "an," and / or "the" do not specifically refer to the singular and may include the plural. The terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Without further limitations, an element defined by the phrase "comprising an..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes said element.

[0043] It should also be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Unless otherwise expressly specified and limited, the terms "installed," "connected," "linked," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood according to the specific circumstances.

[0044] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A mild cognitive impairment detection system based on multi-level data, characterized in that, include: Terminal devices used to collect user physiological monitoring data, clinical narrative data, and multimodal test data; A data processing module for performing interference identification and information purification on the multimodal test data to generate a purified cognitive response matrix; A cognitive assessment module for decoupling cognitive abilities and benchmarking against groups based on the purified cognitive response matrix, and outputting individual ability levels. This is a closed-loop management module used to construct a longitudinal decline slope model based on a scoring sequence of multiple historical assessments of an individual, and to provide risk warnings and intervention recommendations.

2. The system according to claim 1, characterized in that, The terminal devices include smart bracelet terminals, mobile phone terminals, and multimodal testing equipment; The smart bracelet terminal is used to collect users' physiological monitoring data, including: sleep stage data, motor behavior data, and autonomic nervous system data; The mobile terminal is used to acquire clinical narrative data, including: medication records, historical diagnosis results, and travel history; The multimodal testing equipment includes an embedded industrial computer, a microphone array, an eye tracker, a touch screen, and a camera, used to perform multimodal cognitive tests and collect multimodal test data.

3. The system according to claim 1, characterized in that, Interference identification of the multimodal test data includes: Interference is identified through a pre-set causal rule base, which defines the mapping relationship between interference keywords and data operation instructions, and is used to match interference keywords with user physiological monitoring data and clinical narrative data. When a match is successful, execute the data manipulation instructions corresponding to the interference keywords to freeze or weight specific fields in the multimodal test data.

4. The system according to claim 3, characterized in that, The multimodal test data is subjected to information purification to generate a purified cognitive response matrix, including: Determine the sensitivity score of each test item to interference from the interference source, and remove test data of test items that are susceptible to interference from the current interference source based on the sensitivity score; The test items for the same cognitive dimension are tested repeatedly with slight modifications, and the response consistency coefficient is calculated. If the response consistency coefficient is lower than a set value, the test data for the corresponding test item is removed. The individualized interference suppression network dynamically generates fusion weights for each modality test data; and the fusion weights are used to weight and fuse the test data of each modality to obtain a purified cognitive response matrix.

5. The system according to claim 4, characterized in that, The cognitive assessment module includes: a multi-head cognitive decoupling module; The multi-head cognitive decoupling module processes the purified cognitive response matrix through a neural network and outputs scores for multiple individual abilities respectively. The aforementioned individual abilities include: working memory, attention, language ability, executive function, visuospatial ability, and orientation.

6. The system according to claim 5, characterized in that, After outputting scores for multiple individual abilities, the following is also included: The score for each individual ability is compared with the reference distribution of the same age, education level, and gender group to obtain a score reflecting the distribution level of individual abilities.

7. The system according to claim 6, characterized in that, The longitudinal decline slope model is a linear mixed-effects model; The linear mixed-effects model estimates the fixed-effects slope and random-effects slope of an individual's cognitive ability by fitting a series of ratings from multiple historical assessments. The closed-loop management module calculates an individualized slope value based on the fixed-effects slope and the random-effects slope. If the individualized slope value is less than a set threshold, a recession risk warning is issued.

8. The system according to claim 7, characterized in that, The closed-loop management module also includes: intervention recommendation engine; The intervention recommendation engine determines intervention measures and recommends them to users based on the individualized slope value, user physiological monitoring data, and clinical narrative data.

Images