Cognitive impairment assessment device and method based on multi-sensor and machine learning

Abstract

The application discloses a cognitive impairment evaluation device and method based on multiple sensors and machine learning, relates to the technical field of information processing, and comprises a data acquisition module, which is used for synchronously collecting multi-dimensional motion signals generated by a target object in a dynamic balance test process through multiple sensors; a feature screening module, which is used for constructing a feature subset according to the multi-dimensional motion signals; and a cognitive evaluation module, which is used for sequentially performing corresponding cognitive state evaluation on the feature subset through each cognitive impairment evaluation submodel of a preset cognitive impairment evaluation model, predicting and outputting a prediction result of the target object being in a target cognitive state; wherein the cognitive impairment evaluation submodel comprises a health-Parkinson disease cognitive normal binary classification model, a Parkinson disease cognitive normal-Parkinson disease mild cognitive impairment binary classification model and a Parkinson disease mild cognitive impairment-Parkinson disease dementia binary classification model. The three submodels enable the device to distinguish which stage the target object is in.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to a cognitive impairment assessment device and method based on multi-sensor and machine learning. Background Technology

[0002] Parkinson's disease is a common neurodegenerative disease. Besides typical motor symptoms, cognitive impairment is also a significant non-motor symptom. There is a close correlation between cognitive decline and postural balance disorders. Currently, clinical assessment of cognitive and balance functions in Parkinson's patients mainly relies on traditional clinical scales such as the Montreal Cognitive Assessment Scale, the Mini-Mental State Examination, and the Berger Balance Scale. While these methods are easy to use, they have limitations in reflecting subtle early functional changes, revealing the dynamic processes of motor execution, and assessing cognitive-motor integration. Scale scores often concentrate in the higher range, making it difficult to reflect differences in patients with early or mild cognitive states. Furthermore, they are primarily outcome-based, failing to quantify dynamic information such as postural adjustments, continuity, and stability during motor execution.

[0003] However, there is a lack of a systematic approach in the current technology that can effectively integrate functional dynamic equilibrium tasks with high-dimensional kinematic data and use machine learning techniques to construct an objective and quantitative assessment scheme that can sensitively distinguish different cognitive stages of Parkinson's disease and can be used for early identification. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a cognitive impairment assessment device and method based on multi-sensor and machine learning, capable of fully automated staging assessment of the cognitive state of Parkinson's disease patients. The specific solution is as follows: In a first aspect, this application discloses a cognitive impairment assessment device based on multi-sensor and machine learning, applied to a computer device, comprising: The data acquisition module is used to simultaneously acquire multidimensional motion signals generated by the target object during the dynamic balance test through multiple sensors; The feature filtering module is used to construct a feature subset based on the multidimensional motion signal; The cognitive assessment module is used to sequentially assess the corresponding cognitive state of the feature subset through each cognitive impairment assessment sub-model of the preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in the target cognitive state; wherein, the cognitive impairment assessment sub-model includes a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model.

[0005] Optionally, the cognitive impairment assessment device based on multi-sensor and machine learning further includes: multiple wearable inertial measurement units worn on multiple preset parts of the target object respectively; Accordingly, the data acquisition module includes: The data acquisition submodule is used to synchronously acquire multidimensional motion signals generated by the target object under the timed standing-up walking test and the four-step grid test through the multiple wearable inertial measurement units.

[0006] Optionally, the data acquisition submodule includes: The first data acquisition unit is used to synchronously acquire the first signal segment generated by the target object under the timing stand-up walking test through the multiple wearable inertial measurement units; The first data extraction unit is used to extract a first set of motion features, including task time, gait spatiotemporal parameters, turning features, and trunk and limb kinematic parameters, from the first signal segment to construct a first multidimensional motion signal. The second data acquisition unit is used to synchronously acquire the second signal segment generated by the target object under the four-step grid test through the multiple wearable inertial measurement units; The second data extraction unit is used to extract a second set of motion features from the second signal segment, including multi-directional gait timing parameters, lower limb swing support phase parameters, and limb coordinated movement parameters, in order to construct a second multi-dimensional motion signal.

[0007] Optionally, the feature filtering module includes: The feature filtering unit is used to calculate the discrimination ability score of each motion signal feature in the multidimensional motion signal, and filter the target multidimensional signals according to the sorting order of the discrimination ability scores to construct a feature subset.

[0008] Optionally, the cognitive assessment module includes: The first cognitive assessment unit is used to input the feature subset into the healthy-Parkinson's disease cognitive normal binary classification model of the preset cognitive impairment assessment model to obtain the first assessment result; The first prediction output unit is used to terminate the evaluation and output a health prediction result if the first evaluation result is a health result. The second cognitive assessment unit is used to input the feature subset into the Parkinson's disease cognitive normal-Parkinson's disease mild cognitive impairment binary classification model if the first assessment result is a Parkinson's disease cognitive normal result, so as to obtain a second assessment result. The second prediction output unit is used to terminate the evaluation and output the prediction result of normal cognition in Parkinson's disease if the second evaluation result is a result of normal cognition in Parkinson's disease. The third cognitive assessment unit is used to input the feature subset into the Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model if the second assessment result is Parkinson's disease mild cognitive impairment result, so as to obtain the third assessment result. The third prediction output unit is used to terminate the evaluation and output the prediction result of Parkinson's disease mild cognitive impairment if the third evaluation result is Parkinson's disease mild cognitive impairment. The fourth prediction output unit is used to terminate the evaluation and output the Parkinson's disease dementia prediction result if the third evaluation result is Parkinson's disease dementia.

[0009] Optionally, the cognitive impairment assessment sub-model is a binary classifier built based on a preset supervised machine learning algorithm, which includes any one or more of support vector machines, random forests, or extreme gradient boosting.

[0010] Optionally, the cognitive impairment assessment device based on multi-sensor and machine learning further includes: The contribution analysis module is used to perform feature importance analysis on the prediction results of the preset cognitive impairment assessment model, so as to determine the contribution of each kinematic feature in the output feature subset to the prediction results based on the analysis results.

[0011] Secondly, this application discloses a cognitive impairment assessment method based on multi-sensor and machine learning, applied to a computer device, comprising: Multiple sensors are used to simultaneously collect multidimensional motion signals generated by the target object during the dynamic balance test process; Construct a feature subset based on the multidimensional motion signal; The cognitive state of the feature subset is assessed sequentially by each cognitive impairment assessment sub-model of the preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in the target cognitive state; wherein, the cognitive impairment assessment sub-model includes a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model.

[0012] Thirdly, this application discloses an electronic device, including: Memory, used to store computer programs; A processor is used to execute the computer program to implement the steps of the aforementioned disclosed multi-sensor and machine learning-based cognitive impairment assessment method.

[0013] Fourthly, this application discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the aforementioned disclosed cognitive impairment assessment method based on multi-sensor and machine learning.

[0014] As can be seen, this application discloses a data acquisition module for synchronously acquiring multidimensional motion signals generated by a target object during a dynamic balance test using multiple sensors; a feature selection module for constructing a feature subset based on the multidimensional motion signals; and a cognitive assessment module for sequentially assessing the corresponding cognitive state of the feature subset using various cognitive impairment assessment sub-models of a preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in a target cognitive state; wherein, the cognitive impairment assessment sub-models include a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model. Therefore, it can be seen that by acquiring multidimensional motion signals, dynamic process information such as the target object's posture adjustment, movement continuity, and stability can be captured, which can reflect subtle motion abnormalities. In addition, through feature screening, the most sensitive indicators to cognitive state can be automatically identified. Furthermore, by constructing a binary classification model of adjacent stages, corresponding to the developmental paths of health, early stage, mild impairment, and dementia, it is more clinically targeted and biologically reasonable than a single multi-classification model or a model that only distinguishes between the target object and healthy people. The three sub-models enable the device to distinguish which stage the target object is in. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the structure of a cognitive impairment assessment device based on multi-sensor and machine learning disclosed in this application; Figure 2 This is a flowchart of a cognitive impairment assessment method based on multi-sensor and machine learning disclosed in this application; Figure 3 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0018] Parkinson's disease (PD) is a common neurodegenerative disease. Besides typical motor symptoms, cognitive impairment is also a significant non-motor symptom. Currently, clinical assessment of cognitive function in PD patients mainly relies on traditional clinical scales such as the Montreal Cognitive Assessment Scale and the Mini-Mental State Examination. While these methods are easy to use, they have limitations in reflecting subtle early functional changes and revealing the dynamic characteristics of motor execution. Scale scores often concentrate in the higher range, making it difficult to reflect differences in patients with early or mild cognitive states. Furthermore, they are primarily outcome-based, failing to quantify the dynamic information during motor execution.

[0019] In recent years, wearable inertial measurement unit (IMU) sensors have been increasingly applied to the functional assessment of Parkinson's disease due to their ability to objectively and continuously acquire high-resolution kinematic data. Meanwhile, dynamic balance tasks such as the TUG test and FSST have proven effective in revealing motor control abnormalities in patients under cognitive load. Based on this, existing technologies have developed solutions that utilize sensors to collect motion data and combine it with machine learning to build classification models to distinguish Parkinson's disease patients from healthy individuals.

[0020] However, existing technologies still have the following shortcomings: First, the assessment granularity is coarse and cannot meet the needs of staging diagnosis and treatment. Current solutions mostly focus on building a single binary classification model (patient vs. healthy person), which can only answer the question of whether a disease exists. But in clinical practice, doctors need to clearly define whether a patient is cognitively normal, has mild cognitive impairment, or is in the dementia stage in order to develop precise intervention strategies. Current technologies lack a solution that can cover the entire disease course and distinguish between adjacent cognitive stages.

[0021] Second, the assessment process relies on manual intervention and lacks sufficient automation. Even with multiple models for different stages, current technology typically requires doctors to pre-determine the patient's likely stage based on clinical experience and then manually select the appropriate model for verification. This semi-automatic assessment process not only increases the workload for doctors but also heavily depends on clinical experience for accuracy, failing to automate the process from data collection to staging results.

[0022] Third, the ability to identify early subtle changes is insufficient. Current technology has limited ability to capture early subtle changes in patients transitioning from a normal state to mild impairment, which can easily lead to missed diagnoses or delayed intervention.

[0023] Therefore, this application provides a cognitive impairment assessment scheme based on multi-sensor and machine learning, which can automatically and sensitively identify different cognitive stages of Parkinson's disease patients, especially the objective quantitative assessment of early subtle changes.

[0024] like Figure 1 As shown, the present invention provides a cognitive impairment assessment device based on multi-sensor and machine learning, applied to a computer device, comprising: Data acquisition module 11 is used to synchronously acquire multidimensional motion signals generated by the target object during the dynamic balance test through multiple sensors; Feature filtering module 12 is used to construct a feature subset based on the multidimensional motion signal; The cognitive assessment module 13 is used to sequentially assess the corresponding cognitive state of the feature subset through each cognitive impairment assessment sub-model of the preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in the target cognitive state; wherein, the cognitive impairment assessment sub-model includes a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model.

[0025] It is understood that the cognitive impairment assessment device based on multi-sensor and machine learning further includes: multiple wearable inertial measurement units (IMUs) worn on multiple preset sites on the target object; correspondingly, the data acquisition module 11 includes: a data acquisition submodule, used to synchronously acquire multidimensional motion signals generated by the target object under the timed up and go (TUG) test and the four-square step test through the multiple wearable IMUs. Specifically, multiple sensors are worn on 10 sensor sites on the back of both wrists, sternum, fifth lumbar vertebra, above the knees of both thighs, below the knees of both calves, and on the back of the metatarsals of both feet. The sensors include inertial measurement units (IMUs), each IMU integrating a three-axis accelerometer and a three-axis gyroscope, with a sampling frequency of 100Hz. The target object is instructed to complete the Timed Up and Go (TUG) test and the Four Square Step Test (FSST) test in sequence, and the multidimensional motion signals of the target object during the two test processes are acquired synchronously. Furthermore, the acquired multidimensional motion signals undergo preprocessing. First, the raw IMU signal is checked for integrity, and segments with missing or obviously abnormal signals are removed. Interpolation is used to correct signals with transient gaps. Second, the signal is filtered, specifically bandpass filtered (e.g., using a first-order Butterworth low-pass filter with a cutoff frequency of 5Hz) to remove baseline drift and high-frequency noise. Then, based on the start / end markers of the test actions (which can be automatically identified by software or manually labeled), valid data segments corresponding to the TUG and FSST tasks are segmented from the continuous signal. Each data segment is then standardized to eliminate inter-individual dimensional differences.

[0026] Specifically, the data acquisition submodule includes: a first data acquisition unit, used to synchronously acquire a first signal segment generated by the target object under the timed stand-up walking test using the multiple wearable inertial measurement units; a first data extraction unit, used to extract a first set of motion features, including task time, gait spatiotemporal parameters, turning characteristics, and trunk and limb kinematic parameters, from the first signal segment to construct a first multidimensional motion signal; a second data acquisition unit, used to synchronously acquire a second signal segment generated by the target object under the four-step grid test using the multiple wearable inertial measurement units; and a second data extraction unit, used to extract a second set of motion features, including multi-directional gait timing parameters, lower limb swing support period parameters, and limb coordinated movement parameters, from the second signal segment to construct a second multidimensional motion signal. It can be understood that the first data acquisition unit is used to synchronously acquire the first signal segment generated by the target object under the timed stand-up walking test using multiple wearable inertial measurement units. The timed stand-up walking test requires the target object to stand up from a chair with armrests, walk 5 meters in a straight line at a preset speed, turn 180 degrees, return to the chair, and sit down. During the test, inertial measurement units (IMUs) worn on 10 sites on the subject's wrists, sternum, lumbar spine, thighs, calves, and arches recorded triaxial acceleration and angular velocity data in real time, with a sampling frequency selectable at 100Hz. A first data extraction unit connected to the first data acquisition unit was used to extract a first set of motion features from the first signal segment. This feature set encompassed the total task completion time and duration of each stage, gait spatiotemporal parameters (mean, standard deviation, and coefficient of variation of stride length, stride speed, stride frequency, and stride duration), symmetry and coordination parameters (difference between left and right sides, symmetry index), turning features (turning duration, peak angular velocity, and angle range), and kinematic parameters of the trunk and limbs (peak angular velocity, range of motion, and variability of each part), as well as frequency domain features (average frequency, median frequency, and band power of the lower limb acceleration / angular velocity signal), totaling over 190 features that could be extracted. The second data acquisition unit quantifies the motion performance of the target object during continuous movement and turning control. It simultaneously acquires the second signal segment generated by the target object under a four-step grid test using multiple wearable inertial measurement units. The four-step grid test involves setting up four adjacent grids on the ground. Subjects are required to quickly complete forward, lateral, and backward steps in sequence, always facing forward and without touching the grid lines. This test involves multi-directional gait switching and spatial planning, effectively stimulating abnormal motion control under cognitive load. The second data extraction unit is connected to the second data acquisition unit and is used to extract the second motion feature set from the second signal segment.This feature set includes multi-directional gait temporal parameters, such as: total completion time and duration of each directional transition phase; multi-directional gait temporal parameters and their variability; lower limb swing and stance phase parameters; limb coordinated movement angles and angular velocities; frequency domain features (frequency and energy distribution of acceleration / angular velocity signals at key locations), etc., totaling over 570 features that can be extracted. These are used to quantify the coordination and stability of the target object during complex directional changes. Thus, based on two different tasks, two different initial high-dimensional feature sets (a first multi-dimensional motion signal and a second multi-dimensional motion signal) are formed.

[0027] The feature selection module includes a feature selection unit, used to calculate the discriminative ability score of each motion signal feature in the multidimensional motion signal, and to select target multidimensional signals according to the order of the discriminative ability scores to construct a feature subset. It can be understood that each motion signal feature in the first and second multidimensional motion signals is standardized (Z-score), and then the multivariate Sure Independence Screening method is used to evaluate the discriminative ability of each feature for the target classification label. An importance score is obtained by calculating the weighted deviation between the conditional empirical distribution of each feature under different categories and the overall empirical distribution, and all features are sorted in descending order based on the scores. The top N features are selected to form the final feature subset used for modeling. N is an adjustable parameter, preferably between 5 and 20 depending on the actual situation, for example, selecting the top 20 features.

[0028] The cognitive impairment assessment sub-model is a binary classifier built based on a pre-defined supervised machine learning algorithm, which includes any one or more of Support Vector Machines, Random Forests, or Extreme Gradient Boosting. Therefore, the classification tasks in this embodiment are defined as three core binary classification tasks, corresponding to the natural progression path of Parkinson's disease cognitive impairment: Task 1: Healthy Control (HC) vs. Parkinson's Disease with Normal Cognition (PD-NC); Task 2: Parkinson's Disease with Normal Cognition (PD-NC) vs. Parkinson's Disease with Mild Cognitive Impairment (PD-MCI); Task 3: Parkinson's Disease with Mild Cognitive Impairment (PD-MCI) vs. Parkinson's Disease with Dementia (PDD). For each classification task, a selection of corresponding feature subsets is used to model the model using various supervised learning algorithms, including Support Vector Machines, Random Forests, and Extreme Gradient Boosting. On the training set, a hierarchical five-fold cross-validation strategy is used to train the model, and a grid search method is used to optimize the key hyperparameters of each algorithm (such as the kernel function and penalty coefficient C of SVM, and the number of trees and maximum depth of RF). Model performance is validated through cross-validation and independent test sets. Evaluation metrics include accuracy, sensitivity, specificity, F1 score, and area under the receiver operating characteristic (AUC) curve. The model and its parameters that perform best in cross-validation are selected, and the entire training set is used for final training to obtain a classification model suitable for practical evaluation.

[0029] The cognitive assessment module includes: a first cognitive assessment unit, configured to input the feature subset into the healthy-Parkinson's disease cognitively normal binary classification model of the preset cognitive impairment assessment model to obtain a first assessment result; a first prediction output unit, configured to terminate the assessment and output a health prediction result if the first assessment result is a healthy result; a second cognitive assessment unit, configured to continue inputting the feature subset into the Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model if the first assessment result is a Parkinson's disease cognitively normal result to obtain a second assessment result; and a second prediction output unit, configured to output a second assessment result if the second assessment result is a Parkinson's disease cognitively normal result. If the second assessment result indicates mild cognitive impairment in Parkinson's disease, the evaluation is terminated and a prediction result of normal cognitive impairment in Parkinson's disease is output. A third cognitive assessment unit, if the second assessment result indicates mild cognitive impairment in Parkinson's disease, inputs the feature subset into the Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model to obtain a third assessment result. A third prediction output unit, if the third assessment result indicates mild cognitive impairment in Parkinson's disease, terminates the evaluation and outputs a prediction result of mild cognitive impairment in Parkinson's disease. A fourth prediction output unit, if the third assessment result indicates Parkinson's disease dementia, terminates the evaluation and outputs a prediction result of Parkinson's disease dementia. It is understood that the first cognitive assessment unit is connected to the feature filtering module and is used to input the filtered feature subset into the healthy-Parkinson's disease normal cognitive impairment binary classification model in the preset cognitive impairment assessment model. After optimization with training set data, it has the ability to distinguish between healthy individuals and Parkinson's disease patients with normal cognitive impairment. The first cognitive assessment unit runs this model and outputs a first assessment result, which is usually presented in the form of a probability of belonging to a certain category or a hard classification label. The first prediction output unit is connected to the first cognitive assessment unit and is used to judge the first assessment result. If the model output is a healthy result (i.e., classified as a healthy control), it indicates that the target subject does not exhibit Parkinson's disease-related motor abnormalities and has a normal cognitive state. At this time, the subsequent assessment process is terminated, and the healthy prediction result is directly output. If the first assessment result is a Parkinson's disease cognitively normal result, it indicates that the target subject has Parkinson's disease but cognitive function has not yet been impaired, and further assessment is needed to determine whether their cognitive state has evolved. At this time, the second cognitive assessment unit is activated to continue inputting the same feature subset into the Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model to obtain the second assessment result. The second prediction output unit is connected to the second cognitive assessment unit and is used to judge the second assessment result. If the model output is a Parkinson's disease cognitively normal result, it confirms that the target subject is still in the cognitively normal stage, the assessment is terminated, and the Parkinson's disease cognitively normal prediction result is output; if the output is a Parkinson's disease mild cognitive impairment result, it indicates that early cognitive decline has occurred, and the next stage of assessment is required.When the second assessment result is mild cognitive impairment (MCI) of Parkinson's disease, the third cognitive assessment unit is activated to input a subset of features into the MCI-Dementia binary classification model to obtain the third assessment result. This model is specifically designed for the critical stage of transition from mild cognitive impairment to dementia and has high sensitivity in identifying late-stage cognitive decline. The third prediction output unit is connected to the third cognitive assessment unit and is used to judge the third assessment result. If the model output is MCI-Dementia, the assessment is terminated and a MCI-Dementia prediction result is output; if the output is Dementia, the fourth prediction output unit terminates the assessment and outputs a Dementia-Parkinson's disease prediction result. Through the sequential collaboration of the above units, the cognitive assessment module achieves fully automated reasoning from the initial feature input to the final cognitive staging. The entire process does not require clinicians to pre-determine the patient's stage or manually select a model, improving assessment efficiency and objectivity. Meanwhile, this hierarchical assessment strategy is highly consistent with the natural evolutionary path of cognitive impairment in Parkinson's disease, and can sensitively capture the continuous changes from normal to mild impairment and from mild impairment to dementia, providing accurate and interpretable staging diagnostic basis for clinical practice.

[0030] In this embodiment, the device further includes a contribution analysis module, used to perform feature importance analysis on the prediction results of a preset cognitive impairment assessment model, so as to determine the contribution of each kinematic feature in the output feature subset to the prediction results based on the analysis results. It is understood that the contribution analysis module uses the SHAP method to perform interpretability analysis on the model prediction results. Specifically, for each input sample to be evaluated, the contribution analysis module calculates the SHAP value of each kinematic feature in the feature subset. This value quantifies the marginal contribution of the feature to the model output results. A positive SHAP value indicates that the feature pushes the prediction results towards the positive category (cognitive impairment direction), while a negative value indicates that it pushes them towards the negative category. The larger the absolute value of the feature's SHAP value, the higher its contribution to the current prediction result. Thus, the contribution analysis module can output a feature importance ranking chart or a feature contribution distribution chart for individual samples, intuitively displaying the key kinematic indicators affecting the model's decision-making. For example, when distinguishing between normal cognition and mild cognitive impairment in Parkinson's disease, the analysis results can reveal that features such as increased gait variability and decreased motor fluency contribute most significantly to the model's predictions.

[0031] As can be seen, this application discloses a data acquisition module for synchronously acquiring multidimensional motion signals generated by a target object during a dynamic balance test using multiple sensors; a feature selection module for constructing a feature subset based on the multidimensional motion signals; and a cognitive assessment module for sequentially assessing the corresponding cognitive state of the feature subset using various cognitive impairment assessment sub-models of a preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in a target cognitive state; wherein, the cognitive impairment assessment sub-models include a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model. Therefore, it can be seen that by acquiring multidimensional motion signals, dynamic process information such as the target object's posture adjustment, movement continuity, and stability can be captured, which can reflect subtle motion abnormalities. In addition, through feature screening, the most sensitive indicators to cognitive state can be automatically identified. Furthermore, by constructing a binary classification model of adjacent stages, corresponding to the developmental paths of health, early stage, mild impairment, and dementia, it is more clinically targeted and biologically reasonable than a single multi-classification model or a model that only distinguishes between the target object and healthy people. The three sub-models enable the device to distinguish which stage the target object is in.

[0032] Validation data from 138 subjects (including HC, PD-NC, PD-MCI, and PDD) showed that kinematic features are correlated with cognitive function: completion times of FSST and TUG were significantly negatively correlated with total scores on the MoCA and PD-CRS cognitive scales (|r| > 0.4, p < 0.001). Multiple kinematic features (such as gait speed, stride length, turn rate, and pause time during tasks) showed moderate to high correlations with cognitive scores, confirming that the extracted features effectively reflect cognitive function levels. The classification model performed excellently: the constructed machine learning model demonstrated outstanding performance in distinguishing different cognitive states. For example, on the independent test set, the model's AUC for distinguishing between HC and PD reached 0.977, the AUC for distinguishing between PD-NC and PD-MCI reached 0.793, and the AUC for distinguishing between PD-MCI and PDD reached 0.821. These performances significantly outperformed the traditional Berg Balance Scale (BBS) in the corresponding tasks (AUC approximately 0.654). The model's interpretability reveals patterns in pathological evolution: SHAP interpretability analysis showed that early cognitive changes (PD-NC vs. PD-MCI) are more associated with increased gait variability and decreased motor fluency; while late-stage dementia transformation (PD-MCI vs. PDD) is more associated with loss of coordination and planning errors in complex tasks. This confirms that the behavioral pattern changes captured by the model are consistent with theories of pathological progression of cognitive impairment.

[0033] Sample and equipment, study subjects: A total of 138 people were included, including 34 healthy controls (HC), 42 people with normal cognitive function in Parkinson's disease (PD-NC), 41 people with mild cognitive impairment in Parkinson's disease (PD-MCI), and 21 people with dementia in Parkinson's disease (PDD).

[0034] Sensor equipment: The MATRIX wearable system is used, which includes 10 IMU nodes, a sampling rate of 100Hz, an accelerometer range of ±8g, a sensitivity of 4096LSB / g, a gyroscope range of ±2000 dps, and a sensitivity of 16.4LSB / dps.

[0035] 191 features, such as "average duration of 180° turn", "average stride speed", and "standard deviation of gait cycle", were extracted from the TUG task.

[0036] 578 features were extracted from the FSST task, such as "total completion time", "average time spent in the same quadrant with both feet", and "standard deviation of stride width".

[0037] The MV-SIS method was used to select the top 20 most important features from the feature combinations of the two tasks to build a “PD-NC vs PD-MCI” classification model.

[0038] For the classification task of "PD-MCI vs PDD", SVM (using radial basis kernel function, C=10), RF (number of decision trees=100, maximum depth=10), XGBoost (learning rate=0.1, maximum depth=6) and corresponding hyperparameter combinations were used for modeling.

[0039] After optimization through five-fold cross-validation, the XGBoost model with the highest AUC on this task was finally selected as the final model, with an AUC of 0.821 and an accuracy of 78.6% on the independent test set.

[0040] like Figure 2 As shown, the present invention also discloses a cognitive impairment assessment method based on multi-sensor and machine learning, applied to a computer device, comprising: Step S11: Simultaneously acquire multidimensional motion signals generated by the target object during the dynamic balance test using multiple sensors.

[0041] Step S12: Construct a feature subset based on the multidimensional motion signal.

[0042] Step S13: The cognitive state of the feature subset is evaluated sequentially by each cognitive impairment assessment sub-model of the preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in the target cognitive state; wherein, the cognitive impairment assessment sub-model includes a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model.

[0043] The detailed processes in steps S11 to S13 are described in the aforementioned embodiments and will not be repeated here.

[0044] By integrating a dynamic equilibrium task paradigm and utilizing wearable sensors to acquire motion data, this invention overcomes the shortcomings of traditional scales, such as high subjectivity and low resolution. Through high-dimensional feature extraction and feature selection, it can capture key features closely related to cognitive function from massive amounts of kinematic information. Furthermore, by constructing a machine learning classification model targeting the natural progression stages of the disease (adjacent cognitive stages), the method of this invention can not only effectively distinguish between Parkinson's disease patients and healthy individuals, but more importantly, it can sensitively identify subtle changes in cognitive state (from normal cognition to mild cognitive impairment) in the early stages of the disease. This enables objective, quantitative, and staging assessment of cognitive impairment in Parkinson's disease, providing a new technical means for early clinical diagnosis and intervention.

[0045] Furthermore, embodiments of this application also disclose an electronic device, Figure 3 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application.

[0046] Figure 3 This is a schematic diagram of the structure of an electronic device 20 provided in an embodiment of this application. Specifically, the electronic device 20 may include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the cognitive impairment assessment method based on multi-sensor and machine learning disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0047] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0048] The processor 21 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 21 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 21 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 21 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 21 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0049] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0050] The operating system 221 manages and controls the various hardware devices and computer programs 222 on the electronic device 20 to enable the processor 21 to perform calculations and processing on the massive amounts of data 223 in the memory 22. It can be Windows Server, Netware, Unix, Linux, etc. The computer program 222, in addition to including a computer program capable of performing the multi-sensor and machine learning-based cognitive impairment assessment method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, may further include computer programs capable of performing other specific tasks. The data 223 may include data received by the electronic device from external devices, as well as data collected by its own input / output interface 25.

[0051] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned cognitive impairment assessment method based on multi-sensor and machine learning. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0052] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0053] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application. The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly in hardware, software modules executed by a processor, or a combination of both. The software module may be located in random access memory (RAM), memory, read-only memory (ROM), electrically programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, removable disks, CD-ROMs (Compact Disc-Read Only Memory), or any other form of storage medium known in the art.

[0054] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, 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, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0055] The solution provided by the present invention has been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only for helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A cognitive impairment assessment device based on multi-sensor and machine learning, characterized in that, Applied to computer devices, including: The data acquisition module is used to simultaneously acquire multidimensional motion signals generated by the target object during the dynamic balance test through multiple sensors; The feature filtering module is used to construct a feature subset based on the multidimensional motion signal; The cognitive assessment module is used to sequentially assess the corresponding cognitive state of the feature subset through each cognitive impairment assessment sub-model of the preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in the target cognitive state; wherein, the cognitive impairment assessment sub-model includes a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model.

2. The cognitive impairment assessment device based on multi-sensor and machine learning according to claim 1, characterized in that, Also includes: Multiple wearable inertial measurement units are worn on multiple preset parts of the target object; Accordingly, the data acquisition module includes: The data acquisition submodule is used to synchronously acquire multidimensional motion signals generated by the target object under the timed standing-up walking test and the four-step grid test through the multiple wearable inertial measurement units.

3. The cognitive impairment assessment device based on multi-sensor and machine learning according to claim 2, characterized in that, The data acquisition submodule includes: The first data acquisition unit is used to synchronously acquire the first signal segment generated by the target object under the timing stand-up walking test through the multiple wearable inertial measurement units; The first data extraction unit is used to extract a first set of motion features, including task time, gait spatiotemporal parameters, turning features, and trunk and limb kinematic parameters, from the first signal segment to construct a first multidimensional motion signal. The second data acquisition unit is used to synchronously acquire the second signal segment generated by the target object under the four-step grid test through the multiple wearable inertial measurement units; The second data extraction unit is used to extract a second set of motion features from the second signal segment, including multi-directional gait timing parameters, lower limb swing support phase parameters, and limb coordinated movement parameters, in order to construct a second multi-dimensional motion signal.

4. The cognitive impairment assessment device based on multi-sensor and machine learning according to claim 1, characterized in that, The feature filtering module includes: The feature filtering unit is used to calculate the discrimination ability score of each motion signal feature in the multidimensional motion signal, and filter the target multidimensional signals according to the sorting order of the discrimination ability scores to construct a feature subset.

5. The cognitive impairment assessment device based on multi-sensor and machine learning according to claim 1, characterized in that, The cognitive assessment module includes: The first cognitive assessment unit is used to input the feature subset into the healthy-Parkinson's disease cognitive normal binary classification model of the preset cognitive impairment assessment model to obtain the first assessment result; The first prediction output unit is used to terminate the evaluation and output a health prediction result if the first evaluation result is a health result. The second cognitive assessment unit is used to input the feature subset into the Parkinson's disease normal cognitive-Parkinson's disease mild cognitive impairment binary classification model if the first assessment result is a Parkinson's disease cognitive normal result, so as to obtain a second assessment result. The second prediction output unit is used to terminate the evaluation and output the prediction result of normal cognition in Parkinson's disease if the second evaluation result is a result of normal cognition in Parkinson's disease. The third cognitive assessment unit is used to input the feature subset into the Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model if the second assessment result is Parkinson's disease mild cognitive impairment result, so as to obtain the third assessment result. The third prediction output unit is used to terminate the evaluation and output the prediction result of Parkinson's disease mild cognitive impairment if the third evaluation result is Parkinson's disease mild cognitive impairment. The fourth prediction output unit is used to terminate the evaluation and output the Parkinson's disease dementia prediction result if the third evaluation result is Parkinson's disease dementia.

6. The cognitive impairment assessment device based on multi-sensor and machine learning according to claim 1, characterized in that, The cognitive impairment assessment sub-model is a binary classifier built based on a pre-defined supervised machine learning algorithm, which includes any one or more of support vector machines, random forests, or extreme gradient boosting.

7. The cognitive impairment assessment device based on multi-sensor and machine learning according to any one of claims 1 to 6, characterized in that, Also includes: The contribution analysis module is used to perform feature importance analysis on the prediction results of the preset cognitive impairment assessment model, so as to determine the contribution of each kinematic feature in the output feature subset to the prediction results based on the analysis results.

8. A cognitive impairment assessment method based on multi-sensor and machine learning, characterized in that, Applied to computer devices, including: Multiple sensors are used to simultaneously collect multidimensional motion signals generated by the target object during the dynamic balance test process; Construct a feature subset based on the multidimensional motion signal; The cognitive state of the feature subset is assessed sequentially by each cognitive impairment assessment sub-model of the preset cognitive impairment assessment model, so as to predict and output the prediction result of the target object being in the target cognitive state; wherein, the cognitive impairment assessment sub-model includes a healthy-Parkinson's disease cognitively normal binary classification model, a Parkinson's disease cognitively normal-Parkinson's disease mild cognitive impairment binary classification model, and a Parkinson's disease mild cognitive impairment-Parkinson's disease dementia binary classification model.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the steps of the cognitive impairment assessment method based on multi-sensor and machine learning as described in claim 8.

10. A computer-readable storage medium, characterized in that, Used to store a computer program; wherein, when the computer program is executed by a processor, it implements the steps of the cognitive impairment assessment method based on multi-sensor and machine learning as described in claim 8.

Images