Cognitive detection support method and system based on stroop paradigm and near-infrared data

Abstract

The application provides a cognitive detection support method and device based on a Stroop paradigm and near-infrared data, which comprises the following steps: taking the coupling strength of HbO signals in the brain region at each frequency band in fNIRS data as a brain effect connection graph, extracting data features of the brain effect connection graph through a convolutional neural network to obtain brain effect connection features; extracting data features of the fNIRS data through the convolutional neural network to obtain spatiotemporal spectrum features; splicing the brain effect connection features, the spatiotemporal spectrum features and data features of a task execution result to obtain fusion features, and taking the probability of a training user belonging to AD, MCI and health as a cognitive detection support result according to the fusion features by a classifier. The application can intelligently fuse various different perspective data and comprehensively evaluate the cognitive state of a user.

Description

Technical Field

[0001] This invention relates to the fields of cognitive assessment, the Stroop paradigm, and artificial intelligence, and particularly to a cognitive detection support method and system based on the Stroop paradigm and near-infrared data. Background Technology

[0002] Alzheimer's disease (AD) is the most common type of dementia in the elderly, seriously threatening their cognitive health and imposing a heavy burden on society, the economy, and families. Because treatment and intervention are not very effective for patients in the middle and late stages of AD, early screening for AD is particularly important. Mild cognitive impairment (MCI) is a transitional stage from normal aging to AD, representing the optimal period for early intervention. MCI refers to mild functional impairments in working memory, attention, language, and spatial abilities; 80% of MCI patients will progress to AD within 6 years. Therefore, analyzing changes in brain connectivity in MCI and AD patients and conducting effective cognitive screening and training are of great significance. Currently, cognitive assessments are mostly conducted using scales such as the Mini-Mental State Examination (MMSE) and the Montreal Cognitive Assessment (MoCA), which have drawbacks such as being time-consuming, requiring professional personnel, and being highly susceptible to subjective factors.

[0003] Stroop is a widely used cognitive assessment paradigm in psychology and clinical medicine, effectively assessing working memory and attention. It is often used to analyze the ability of healthy and Alzheimer's disease (AD) older adults to inhibit cognitive interference. When completing the Stroop task, subjects experience inconsistent cognitive interference, such as different colored text content and text color. Healthy and AD older adults generally have longer reaction times than younger people during the Stroop task, but under inconsistent text and color interference, AD older adults perform worse than healthy older adults. Current technology utilizes functional magnetic resonance imaging (fMRI) to study the neural basis of cognitive assessment tasks such as Stroop and Strooplike. Currently, the changes in brain activation and connectivity in AD patients under Stroop and other cognitive assessment tasks are not fully understood, and the accuracy of cognitive assessment of AD patients based on the Stroop paradigm lacks effective validation.

[0004] In addition, patents (CN202111578471.7, CN202110295106.9, CN201710208847.2, WO2021112987-A1, US2016067244-A1, and WO2022120089-A1) provide intervention methods for cognitive impairment-related diseases, using the Stroop test as one of the cognitive assessment tools, but do not improve how to use the Stroop test. Patent (CN202111186213.4) integrates the Stroop test with multiple other cognitive assessment tests and uses statistical methods to determine the final cognitive status, but does not optimize the Stroop test, integrate neuroimaging data, or use deep learning methods to build the final cognitive assessment model.

[0005] Patents (CN202111578471.7, CN202110295106.9, CN201710208847.2, WO2021112987-A1, US2016067244-A1, WO2022120089-A1 and CN202111186213.4) all use the Stroop test as one of the cognitive assessment methods. However, the above technologies only use the traditional Stroop test and do not improve how to use the Stroop test. They cannot solve the problems of insufficient accuracy of traditional Stroop test assessment alone and the complexity of operation requiring professional physicians. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention proposes a cognitive detection support method based on the Stroop paradigm and near-infrared data, which includes:

[0007] Step 1: Acquire near-infrared fNIRS data of the training user during the Stroop cognitive assessment task, and record the task execution results through a data storage device;

[0008] Step 2: Obtain the coupling strength of HbO signals in brain regions at various frequency bands in the fNIRS data through dynamic Bayesian inference as a brain effect connectivity map. Extract the data features of the brain effect connectivity map through the first convolutional neural network to obtain brain effect connectivity features; extract the data features of the fNIRS data through the second convolutional neural network to obtain spatiotemporal spectral features.

[0009] Step 3: Concatenate the brain effect connectivity feature, the spatiotemporal spectrum feature, and the data feature of the task execution result to obtain the fusion feature. The classifier uses the fusion feature to obtain the probability that the training user belongs to AD, MCI, and healthy as the cognitive detection result. Based on the cognitive detection result and the training user's real cognitive detection label, construct the loss function to train the first convolutional neural network, the second convolutional neural network, and the classifier.

[0010] Step 4: By inputting the brain effect connectivity map and fNIRS data of the user to be tested into the first convolutional neural network and the second convolutional neural network after training, respectively, the brain effect connectivity map and fNIRS data features of the user to be tested are obtained. These features are then concatenated with the Stroop task execution features of the user to be tested and input into the classifier after training to obtain the probability that the user to be tested belongs to AD, MCI, and healthy. This probability serves as the cognitive testing support result for the user to be tested, in order to assist the user or doctor in understanding the cognitive ability of the user to be tested.

[0011] The cognitive detection support method based on the Stroop paradigm and near-infrared data includes the following Stroop task execution characteristics: Stroop task accuracy, error rate, and omission rate.

[0012] The cognitive detection method based on the Stroop paradigm and near-infrared data includes step 1, which involves: presenting a Stroop cognitive assessment task to a training user and recording the training user's task completion status as the task execution result through a data storage device; and synchronously acquiring the training user's fNIRS data through a functional near-infrared device.

[0013] The cognitive detection support method based on the Stroop paradigm and near-infrared data, wherein the process of constructing the brain effect connectivity map in step 2 includes:

[0014] The fNIRS data received by the functional near-infrared device is converted into blood oxygen data. The blood oxygen signal in the specified frequency band is obtained through filtering. The average blood oxygen concentration during the execution of the Stroop cognitive assessment task is calculated and wavelet transform is performed to obtain the signal change in the specified band. The average phase information of each frequency band is calculated using the phase angle function to derive the phase oscillation model between the two channels. The coupling strength of the HbO signal in the brain regions at each frequency is calculated based on dynamic Bayesian inference to establish the brain effect connectivity map.

[0015] This invention also proposes a cognitive detection support device based on the Stroop paradigm and near-infrared data, comprising:

[0016] Module 1: Acquire near-infrared fNIRS data of training users during the Stroop cognitive assessment task, and record the task execution results through a data storage device;

[0017] Module 2: The coupling strength of HbO signal in brain regions at various frequency bands in the fNIRS data is obtained through dynamic Bayesian inference as a brain effect connectivity map. The data features of the brain effect connectivity map are extracted through the first convolutional neural network to obtain brain effect connectivity features. The data features of the fNIRS data are extracted through the second convolutional neural network to obtain spatiotemporal spectral features.

[0018] Module 3: The brain effect connection feature, the spatiotemporal spectrum feature, and the data feature of the task execution result are concatenated to obtain the fusion feature. The classifier uses the fusion feature to obtain the probability that the training user belongs to AD, MCI, and healthy as the cognitive detection result. Based on the cognitive detection result and the training user's real cognitive detection label, a loss function is constructed to train the first convolutional neural network, the second convolutional neural network, and the classifier.

[0019] Module 4: By inputting the brain effect connectivity map and fNIRS data of the user to be tested into the first convolutional neural network and the second convolutional neural network after training, respectively, the brain effect connectivity map and fNIRS data features of the user to be tested are obtained. These features are then concatenated with the Stroop task execution features of the user to be tested and input into the classifier after training to obtain the probability that the user to be tested belongs to AD, MCI, and healthy. This probability serves as the cognitive testing support result for the user to be tested, in order to assist users or doctors in understanding the cognitive abilities of the user to be tested.

[0020] The cognitive detection support device based on the Stroop paradigm and near-infrared data includes the following Stroop task execution characteristics: Stroop task accuracy, error rate, and omission rate.

[0021] The cognitive detection device based on the Stroop paradigm and near-infrared data includes module 1, which comprises: presenting a Stroop cognitive assessment task to a training user and recording the training user's task completion status as the task execution result through a data storage device; and synchronously acquiring the training user's fNIRS data through a functional near-infrared device.

[0022] The cognitive detection support device based on the Stroop paradigm and near-infrared data, wherein the construction process of the brain effect connectivity map in module 2 includes:

[0023] The fNIRS data received by the functional near-infrared device is converted into blood oxygen data. The blood oxygen signal in the specified frequency band is obtained through filtering. The average blood oxygen concentration during the execution of the Stroop cognitive assessment task is calculated and wavelet transform is performed to obtain the signal change in the specified band. The average phase information of each frequency band is calculated using the phase angle function to derive the phase oscillation model between the two channels. The coupling strength of the HbO signal in the brain regions at each frequency is calculated based on dynamic Bayesian inference to establish the brain effect connectivity map.

[0024] The present invention also proposes a server, which includes the aforementioned cognitive detection support device.

[0025] The present invention also proposes a storage medium for storing a computer program that performs the cognitive detection support method.

[0026] As can be seen from the above solutions, the advantages of the present invention are:

[0027] This invention proposes a cognitive assessment method and system based on the Stroop paradigm, aiming to achieve an objective, quantitative, interpretable, and intelligent cognitive state assessment method and system. This method, based on the Stroop paradigm, a portable laptop, and an fNIRS device, achieves objective and quantitative cognitive assessment through simple task design. It can calculate the brain connectivity of subjects under task conditions, visualize and analyze the interaction between different brain regions at different frequencies, thereby providing brain effect connectivity data for the establishment of intelligent cognitive models and revealing the brain function evolution mechanism from normal aging to MCI and AD. This method can effectively assess cognitive state from multiple perspectives. Taking the subject's task completion under the Stroop task, synchronously collected fNIRS data, and brain effect connectivity as inputs, it can comprehensively judge the subject's cognitive state through the intelligent fusion of data from various perspectives. Attached Figure Description

[0028] Figure 1 This is a schematic diagram of the Stroop paradigm based on question and color judgment;

[0029] Figure 2 A schematic diagram of the block design scheme in a Stroop task;

[0030] Figure 3 This is a schematic diagram of a multi-view convolutional neural network;

[0031] Figure 4 A schematic diagram showing the activation of the cerebral cortex in each group of subjects during the Stroop task;

[0032] Figure 5A comparison of the strength of brain connectivity coupling in the resting and task states of subjects in each group under the Stroop paradigm, as detected by the fNIRS device.

[0033] Figure 6 A comparative diagram showing the strength of task-oriented brain connectivity coupling in the Stroop paradigm for each group of subjects, as detected by the fNIRS device. Detailed Implementation

[0034] While researching cognitive assessment methods, the inventors discovered that existing methods based on neuroimaging, neuroelectrophysiology, and scales suffer from high assessment costs, long processing times, the need for professional personnel, and significant susceptibility to subjective factors. A simplified cognitive assessment method is urgently needed. Furthermore, they found that existing deep learning-based cognitive assessment methods often use black-box models to assess and learn cognitive abilities, resulting in poor interpretability. Therefore, an interpretable and visualized approach is urgently needed to visually analyze the brain activation and connectivity states of dementia patients, thereby achieving interpretable cognitive assessment.

[0035] Therefore, this invention proposes a cognitive assessment method and system based on the Stroop paradigm. The method and system design a cognitive assessment task based on the Stroop paradigm, where data is presented and recorded using a portable laptop. Simultaneously, a 61-channel functional near-infrared reflectance spectroscopy (fNIRS) device is used to record near-infrared fNIRS data of the subject during the Stroop task. Specifically, the fNIRS device cap is placed on the user's head, and the signal quality of the 61 electrodes on the cap is detected by an application program in the main control computer. For electrodes with poor signal quality, the hair at each electrode is individually removed to ensure that the near-infrared light can penetrate the scalp and hit the blood vessels in the user's scalp. After adjustment, the near-infrared device signal recording switch is turned on, and the near-infrared signal is recorded to obtain fNIRS data.

[0036] This invention analyzes the Stroop task completion status of subjects and synchronously acquired fNIRS data to analyze the changes in oxygenated hemoglobin (HbO) and deoxygenated hemoglobin (HbR) in the bilateral frontal cortex (PFC), temporal cortex (TC), and occipital cortex (OC) of healthy individuals, those with MCI, and those with AD under resting and task-related conditions. This invention uses a multi-view, multi-scale convolutional neural network to establish a multimodal cognitive assessment model based on Stroop task completion status and fNIRS data to intelligently assess the cognitive state of subjects. The multimodal cognitive assessment model integrates a deep learning model of Stroop task completion status and fNIRS data (including brain connectivity networks and time-frequency features).

[0037] This invention designs a Stroop task, enabling the simultaneous collection of Stroop task results and fNIRS data to construct a deep learning-based cognitive assessment model. This invention proposes a cognitive assessment method and system based on the Stroop paradigm, overcoming the problems of existing cognitive assessment methods such as long processing time, requirement for professional personnel, high subjectivity, and poor interpretability. This invention consists of three parts: Stroop paradigm cognitive assessment task design, brain effect connectivity computation, and intelligent cognitive assessment. The problems addressed by these three parts are as follows:

[0038] 1) Stroop Paradigm Cognitive Assessment Task Design:

[0039] This section primarily overcomes the problems of existing cognitive assessment technologies, such as long processing times, the need for professional personnel, and high subjectivity. The Stroop-based cognitive assessment task requires participants to judge the color or text of red "red," red "blue," blue "red," and blue "blue" words provided in a portable notebook, based on prompts. The task completion process (using a portable notebook) and the participants' fNIRS data during the task are recorded simultaneously. Subsequent brain connectivity calculations and intelligent cognitive assessments are all based on the aforementioned task completion data and fNIRS data.

[0040] 2) Brain effect connectivity calculation:

[0041] This section primarily addresses the issue of poor interpretability in existing cognitive assessments. Based on fNIRS data, this section uses Dynamic Bayesian Inference (DBI) to calculate the coupling strength of HbO signals across different brain regions at different frequencies, thereby obtaining the connectivity of brain effects across different brain regions at different frequency bands.

[0042] 3) Intelligent cognitive assessment:

[0043] This section primarily enables intelligent and automated cognitive assessment. Based on the subject's task completion status under the Stroop task, synchronously collected fNIRS data, and brain effect connectivity, this section establishes a multi-view convolutional neural network to assess the subject's cognitive state.

[0044] To achieve the above-mentioned technical effects, the present invention includes the following key technical points:

[0045] Key Point 1: A method for acquiring cognitive state-related data based on the Stroop paradigm, following the chunking design principle and presenting Stroop tasks in block and trail formats. The Stroop task is presented to the participants using a portable notebook, and the notebook is used to record the participants' task completion status; simultaneously, the participants' fNIRS data are collected to assess changes in their brain state.

[0046] Key Point 2: A brain effect connectivity calculation method based on dynamic Bayesian inference, which uses wavelet transform to calculate fNIRS data in different bands, and uses dynamic Bayesian inference to calculate the effect connectivity in different brain regions based on the fNIRS data in different bands.

[0047] Key Point 3: A method for establishing a cognitive assessment model based on multi-view, multi-scale convolution. The model takes the subject's task completion status under the Stroop task, the synchronously collected fNIRS data, and brain effect connectivity as inputs, and performs intelligent and automated assessment of the subject's cognitive state from different views.

[0048] To make the above-mentioned features and effects of the present invention clearer and easier to understand, specific embodiments are described below in conjunction with the accompanying drawings. This specification discloses one or more embodiments incorporating the features of the present invention. The disclosed embodiments are merely illustrative. The scope of protection of the present invention is not limited to the disclosed embodiments, but is defined by the appended claims.

[0049] 1. Cognitive Assessment Methods and Systems Based on the Stroop Paradigm

[0050] 1.1 Stroop Paradigm Cognitive Assessment Task Design

[0051] The Stroop paradigm is a classic psychological test; a simplified version is illustrated as follows: Figure 1As shown, designed by American psychologist John Ridley Stroop in 1935, it aims to assess subjects' attention, cognitive control, and inhibitory functions by observing their reaction time and accuracy under consistent conditions (e.g., "red" is presented in red), inconsistent conditions (e.g., "red" is presented in green), and neutral conditions (e.g., "good" is presented in green).

[0052] The Stroop task used in this invention requires participants to judge the color or text of given red "red", red "blue", blue "red", and blue "blue" characters based on prompts. The Stroop task process is as follows: Figure 2 As shown, the task design followed the block design principle, presented in block format, consisting of 8 trials (i.e., 32 seconds of task performance followed by 20 seconds of rest). Each trial involved 8 color or text judgments. The task process recorded the subjects' Stroop task color and text judgment results (including accuracy, error rate, and omission rate (i.e., the probability of not judging the task)) and recorded changes in cerebral blood oxygenation using an fNIRS device. Each subject's fNIRS data included resting-state and task-state data. Resting-state data lasted 20 seconds per trial, with 8 repetitions totaling 160 seconds; task-state data lasted 32 seconds per trial, with 8 repetitions totaling 256 seconds.

[0053] 1.2 Brain Effect Connectivity Calculation

[0054] 1.2.1 fNIRS Signal Processing

[0055] (1) Basis of fNIRS signals: fNIRS signals of different frequencies have different physiological origins. Based on the research results, the characteristic frequency bands of fNIRS can be divided as follows: Band I (0.6-2.0 Hz), which is mainly related to heart rate activity; Band II (0.145-0.600 Hz), which is mainly related to respiration; Band III (0.052-0.145 Hz), which is mainly related to myogenic activity; Band IV (0.021-0.052 Hz), which is mainly related to autoneuronic activity; Band V (0.0095-0.0210 Hz), which is mainly related to endothelial cell metabolic activity; and Band VI (0.0050-0.0095 Hz), which is mainly related to endothelial cell activity. This study focuses on brain functional connectivity in frequency bands II, III, and IV for each group of subjects. fNIRS signal processing was implemented in MATLAB R2020a.

[0056] (2) fNIRS signal preprocessing: The fNIRS signal received by the detector is first converted into HbO and HbR according to the modified Lambert-Beer law; then, bandpass filtering is used to remove physiological noise, baseline drift caused by environmental changes and high-frequency noise, etc.; then, moving standard deviation and cubic spline interpolation are used to remove motion artifacts in the data; then, sixth-order Butterworth filtering is used to obtain blood oxygen signals between 0.021 and 2.00 Hz.

[0057] (3) Brain activation mapping based on HbO: The experimental task was designed according to the block principle, with each block containing 10 trails. The fNIRS data could be segmented according to the trails based on the markers to calculate the average blood oxygen concentration of all subjects across all trials. During the brain activation mapping process, the final brain activation map was drawn by averaging the HbO concentration data from 61 sampling channels using time averaging and two-dimensional grid interpolation.

[0058] (4) Brain effect connectivity calculation based on HbO: Wavelet transform is performed on the preprocessed HbO to obtain signal changes in different bands; then, the phase information of each channel is calculated using the phase angle function, and the average phase information is calculated in frequency bands II, III, and IV respectively; then, the phase oscillation model between the two channels is derived, and the coupling strength of the HbO signal between brain regions at different frequencies is calculated based on Dynamic Bayesian Inference (DBI). In addition, it is necessary to verify the true level of brain effect connectivity. The specific process includes: first, calculating the coupling strength between the two brain regions, then repeating 100 times to calculate the substitution signal and the coupling strength between the real signal and the substitution signal; then comparing the coupling strength between the two brain regions with the average of the coupling strength between the 100 substitution signals plus twice the standard deviation to verify whether the coupling strength parameter between the two brain regions is effective.

[0059] 1.2.2 Statistical Processing

[0060] Data processing was performed using SPSS 22.0 statistical software. The data processed included basic information about the subjects and the coupling strength between brain regions. Basic information included age and gender, described using one-dimensional arrays (arrays with dimension 25). The coupling strength between brain regions for each subject was described using 2×3 six-dimensional arrays, where "2" represents the number of states (task state and resting state), "3" represents the number of wavebands (wavebands II, III, and IV), and "6" represents the number of brain regions (left frontal lobe, right frontal lobe, left temporal lobe, right temporal lobe, left occipital lobe, and right occipital lobe). Normality analysis was performed using the Lilliefers test, and homogeneity of variance analysis was performed using the Levene test. Nonparametric one-way ANOVA with the Kruskal-Wallis test was used to test the significance of age, Stroop task results, and coupling strength between brain regions between two and three groups. The chi-square test was used to test the significance of gender. A p-value < 0.05 was considered statistically significant.

[0061] 1.3 Intelligent Cognitive Assessment

[0062] To improve the accuracy and interpretability of the cognitive assessment model, this invention uses three different view data to jointly build the cognitive assessment model, such as... Figure 3 As shown, the causal correlation view and the spatiotemporal spectrum feature view are calculated based on fNIRS data. The causal correlation view represents the brain effect connectivity results, while the spatiotemporal spectrum feature view uses the original fNIRS data as input and learns fNIRS data features through a multi-layer convolutional neural network. The Stroop task-related view uses the accuracy, error rate, and omission rate of the subjects in the Stroop task as input. Finally, the features learned from the three views are concatenated to form a fused feature, which is then input into the classifier to establish an intelligent cognitive assessment model.

[0063] 2. Experimental verification

[0064] 2.1 Experimental Subjects

[0065] The study selected an elderly AD cohort from the Department of Neurology at Foshan First People's Hospital as the research subjects. The subjects were over 55 years old, and the study period was from December 2019 to February 2020. A total of 25 subjects were included, including 9 with normal cognitive function (control group), 4 males and 5 females, with a mean age of (68.00±3.24) years; 10 patients with MCI (MCI group), 5 males and 5 females, with a mean age of (65.70±5.10) years; and 6 patients with AD (AD group), 3 males and 3 females, with a mean age of (65.33±7.17) years. (1) Inclusion criteria for the control group: MMSE ≥ 27 points, MoCA ≥ 26 points, clinical dementiarizing (CDR) score = 0 points; the subject reported normal cognitive function, which was confirmed by an informed person. (2) MCI inclusion criteria: 23 points ≤ MMSE ≤ 26 points, 18 points ≤ MoCA ≤ 25 points, CDR = 0.5; subjects reported their cognitive and daily activities, and it was recommended that a knowledgeable person confirm their cognitive status. Finally, the doctor assessed the subjects based on the subjects' reports or the knowledgeable person's description. (3) AD group inclusion criteria: MMSE ≤ 22 points, MoCA ≤ 17 points, CDR = 1 point. Exclusion criteria: severe hearing and visual impairment; severe neuropsychiatric diseases, neurodegenerative diseases, and other brain diseases, including multiple sclerosis, epilepsy, pain syndrome, restless legs syndrome, stroke, brain injury, traumatic brain injury, cerebral hemorrhage, transient ischemic attack, basal skull fracture, etc.; recent use of psychotropic drugs. All subjects signed written informed consent forms. This study was approved by the Human Ethics Committee of the National Rehabilitation Technology Assisted Research Center and complies with the ethical standards stipulated in the Declaration of Helsinki (2008 revision) in 1975. There were no statistically significant differences in gender composition and age among the three groups (P > 0.05).

[0066] 2.2 Data Collection

[0067] All participants were required to complete the MMSE, MoCA, and CDR scales to assess their overall cognitive abilities, with the results adjusted for their educational background. Memory abilities were assessed using the Auditory-Verbal Learning Test, including immediate and short-term delayed memory; language abilities were assessed using the Boston Naming Test, Symbolic Numerical Pattern Test, and Visuospatial Clock Drawing Test.

[0068] The Stroop task was used to assess participants' cognitive abilities and collect test-related data and fNIRS data during the task. The fNIRS device was a desktop near-infrared acquisition instrument developed by Danyang Huichuang Medical Devices Co., Ltd., capable of recording 61 channels of fNIRS continuous wave changes at a sampling frequency of 17Hz. The laser light source wavelengths were 740nm and 820nm. The fNIRS channels were set according to a 10-20 system, covering six brain regions: left frontal lobe, right frontal lobe, left temporal lobe, right temporal lobe, left occipital lobe, and right occipital lobe. The Stroop task needed to be conducted in a quiet room, and participants were required to remain calm during the task. Before the task began, the task content was explained to the participants in detail, and they were given a certain amount of time to practice until they understood the task rules. After the practice, participants wore near-infrared headgear to detect changes in brain blood oxygenation during the resting state and the Stroop task.

[0069] 2.3 Comparison of Stroop task detection results for each group

[0070] The task completion accuracy rate decreased sequentially among the control group, MCI group, and AD group, while the error rate increased sequentially, and the differences were significant (Table 1).

[0071] Table 1. Task accuracy, error rate, omission rate, and statistical results for each group of subjects.

[0072]

[0073]

[0074] 2.4 Activation of the cerebral cortex in each group of patients

[0075] (1) Compared with the resting state, the activation levels of the bilateral frontotemporal lobes were significantly increased in all three groups of subjects under task conditions; (2) Under task conditions, the activation area in the control group was mainly concentrated in the temporal lobe, and the activation level was relatively high. The activation area in the MCI group was larger, involving both the temporal and frontal lobes, but the activation level was lower than that in the control group. The activation area in the AD group was more extensive, with activation in the temporal, frontal, and occipital lobes, but the activation level was lower. See details below. Figure 4 In the diagram: MCI: Mild cognitive impairment; AD: Alzheimer's disease; different colors represent the degree of cerebral cortex activation, with blue gradually changing to red to indicate that the degree of cerebral cortex activation gradually increases from low to high, and the values ​​in the color bars represent the concentration of oxyhemoglobin.

[0076] 2.5 Comparison of interbrain coupling strength among different patient groups

[0077] The brain region coupling strength of subjects in the control group, MCI group, and AD group under resting and task-oriented conditions is as follows: Figure 5As shown in the figure, in frequency band II, the task-state coupling strength in the LOC→LPFC and LOC→RPFC directions was significantly increased compared to the resting state in the MCI group; in frequency band II, the task-state coupling strength in the RTC→ROC direction was significantly increased compared to the resting state in the AD group; and in frequency band IV, the task-state coupling strength in the ROC→LTC direction was significantly increased compared to the resting state in the control group. The pairwise coupling strengths of the control group, MCI, and AD groups under task-state conditions are shown in the figure. Figure 6 As shown, compared with the MCI group, the AD group showed a significant increase in task-mode coupling strength in frequency band II along the LTC→LPFC and LTC→RTC directions; compared with the MCI group, the control group showed a significant increase in task-mode coupling strength in frequency band III along the ROC→LTC direction; and compared with the MCI group, the AD group showed a significant increase in coupling strength in frequency band III along the LTC→LOC, LTC→ROC, ROC→LTC, and ROC→LOC directions. This indicates that compared with MCI patients, AD patients need to mobilize more brain resources to complete the same task. Figure 5 MCI: Mild cognitive impairment; AD: Alzheimer's disease; fNIRS: Functional near-infrared spectroscopy; LPFC: Left frontal lobe; RPFC: Right frontal lobe; LTC: Left temporal lobe; RTC: Right temporal lobe; LOC: Left occipital lobe; ROC: Right occipital lobe; Warm colors indicate that the coupling strength of the former (HC, HC, and MCI groups) is higher than that of the latter (MCI, AD, and AD groups); ★ indicates whether the difference in coupling strength between the two is statistically significant (P<0.05); Cool colors indicate that the coupling strength of the former is lower than that of the latter; ★ indicates whether the difference in coupling strength between the two is statistically significant (P<0.05); Figure 6 In Chinese: MCI: Mild cognitive impairment; AD: Alzheimer's disease; fNIRS: Functional near-infrared spectroscopy; LPFC: Left frontal lobe; RPFC: Right frontal lobe; LTC: Left temporal lobe; RTC: Right temporal lobe; LOC: Left occipital lobe; ROC: Right occipital lobe; Warm colors indicate that the coupling strength of the former (HC, HC, and MCI groups) is higher than that of the latter (MCI, AD, and AD groups); ★ indicates whether the difference in coupling strength between the two is statistically significant (P<0.05); Cool colors indicate that the coupling strength of the former is lower than that of the latter; ★ indicates whether the difference in coupling strength between the two is statistically significant (P<0.05).

[0078] 2.6 Intelligent Cognitive Assessment

[0079] Table 2 Accuracy of Intelligent Cognitive Assessment Model

[0080]

[0081] In the construction of the intelligent cognitive assessment model, the model was trained using data from 24 out of a total of 25 subjects, and the model's performance was tested using data from the remaining subject. This process was repeated 25 times until data from each subject was used as test data once. The mean performance of the model during the 25 training iterations was statistically analyzed, and the results are shown in Table 2. Experimental results show that the average accuracy rates for the causal association view only, the spatiotemporal spectral feature view only, and the Stroop task-related view only were 65%, 70%, and 60%, respectively, while the multi-view fusion method proposed in this invention achieved an accuracy of 8%, demonstrating the effectiveness of the proposed method.

[0082] The following are system embodiments corresponding to the above method embodiments. This embodiment can be implemented in conjunction with the above embodiments. The relevant technical details mentioned in the above embodiments are still valid in this embodiment, and will not be repeated here to reduce repetition. Accordingly, the relevant technical details mentioned in this embodiment can also be applied to the above embodiments.

[0083] This invention also proposes a cognitive detection support device based on the Stroop paradigm and near-infrared data, comprising:

[0084] Module 1: Acquire near-infrared fNIRS data of training users during the Stroop cognitive assessment task, and record the task execution results through a data storage device;

[0085] Module 2: The coupling strength of HbO signal in brain regions at various frequency bands in the fNIRS data is obtained through dynamic Bayesian inference as a brain effect connectivity map. The data features of the brain effect connectivity map are extracted through the first convolutional neural network to obtain brain effect connectivity features. The data features of the fNIRS data are extracted through the second convolutional neural network to obtain spatiotemporal spectral features.

[0086] Module 3: The brain effect connection feature, the spatiotemporal spectrum feature, and the data feature of the task execution result are concatenated to obtain the fusion feature. The classifier uses the fusion feature to obtain the probability that the training user belongs to AD, MCI, and healthy as the cognitive detection result. Based on the cognitive detection result and the training user's real cognitive detection label, a loss function is constructed to train the first convolutional neural network, the second convolutional neural network, and the classifier.

[0087] Module 4: By inputting the brain effect connectivity map and fNIRS data of the user to be tested into the first convolutional neural network and the second convolutional neural network after training, respectively, the brain effect connectivity map and fNIRS data features of the user to be tested are obtained. These features are then concatenated with the Stroop task execution features of the user to be tested and input into the classifier after training to obtain the probability that the user to be tested belongs to AD, MCI, and healthy. This probability serves as the cognitive testing support result for the user to be tested, in order to assist users or doctors in understanding the cognitive abilities of the user to be tested.

[0088] The cognitive detection support device based on the Stroop paradigm and near-infrared data includes the following Stroop task execution characteristics: Stroop task accuracy, error rate, and omission rate.

[0089] The cognitive detection device based on the Stroop paradigm and near-infrared data includes module 1, which comprises: presenting a Stroop cognitive assessment task to a training user and recording the training user's task completion status as the task execution result through a data storage device; and synchronously acquiring the training user's fNIRS data through a functional near-infrared device.

[0090] The cognitive detection support device based on the Stroop paradigm and near-infrared data, wherein the construction process of the brain effect connectivity map in module 2 includes:

[0091] The fNIRS data received by the functional near-infrared device is converted into blood oxygen data. The blood oxygen signal in the specified frequency band is obtained through filtering. The average blood oxygen concentration during the execution of the Stroop cognitive assessment task is calculated and wavelet transform is performed to obtain the signal change in the specified band. The average phase information of each frequency band is calculated using the phase angle function to derive the phase oscillation model between the two channels. The coupling strength of the HbO signal in the brain regions at each frequency is calculated based on dynamic Bayesian inference to establish the brain effect connectivity map.

[0092] The present invention also proposes a server, which includes the aforementioned cognitive detection support device.

[0093] The present invention also proposes a storage medium for storing a computer program that performs the cognitive detection support method.

[0094] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A cognitive detection support method based on the Stroop paradigm and near-infrared data, characterized in that, include: Step 1: Acquire near-infrared fNIRS data of the training user during the Stroop cognitive assessment task, and record the task execution results through a data storage device; Step 2: Obtain the coupling strength of HbO signals in brain regions at various frequency bands in the fNIRS data through dynamic Bayesian inference as a brain effect connectivity map. Extract the data features of the brain effect connectivity map through the first convolutional neural network to obtain brain effect connectivity features; extract the data features of the fNIRS data through the second convolutional neural network to obtain spatiotemporal spectral features. Step 3: Concatenate the brain effect connectivity feature, the spatiotemporal spectrum feature, and the data feature of the task execution result to obtain the fusion feature. The classifier uses the fusion feature to obtain the probability that the training user belongs to AD, MCI, and healthy as the cognitive detection result. Based on the cognitive detection result and the training user's real cognitive detection label, construct the loss function to train the first convolutional neural network, the second convolutional neural network, and the classifier. Step 4: By inputting the brain effect connectivity map and fNIRS data of the user to be cognitively tested into the first convolutional neural network and the second convolutional neural network after training, respectively, the brain effect connectivity map and fNIRS data features of the user to be cognitively tested are obtained. After concatenation with the Stroop task execution features of the user to be cognitively tested, the data is input into the classifier after training to obtain the probability that the user to be cognitively tested belongs to AD, MCI and healthy, which is used as the cognitive detection support result of the user to be cognitively tested. The process of constructing the brain effect connectivity map in step 2 includes: The fNIRS data received by the functional near-infrared device is converted into blood oxygen data. The blood oxygen signal in the specified frequency band is obtained through filtering. The average blood oxygen concentration during the execution of the Stroop cognitive assessment task is calculated and wavelet transform is performed to obtain the signal change in the specified band. The average phase information of each frequency band is calculated using the phase angle function to derive the phase oscillation model between the two channels. The coupling strength of HbO signal in multiple brain regions at each frequency is calculated based on dynamic Bayesian inference to establish the brain effect connectivity map.

2. The cognitive detection support method based on the Stroop paradigm and near-infrared data as described in claim 1, characterized in that, The execution characteristics of this Stroop task include: Stroop task accuracy, error rate, and omission rate.

3. The cognitive detection support method based on the Stroop paradigm and near-infrared data as described in claim 1, characterized in that, Step 1 includes: presenting the Stroop cognitive assessment task to the training user and recording the training user's task completion status as the task execution result through a data storage device; and synchronously acquiring the training user's fNIRS data through a functional near-infrared device.

4. A cognitive detection support device based on the Stroop paradigm and near-infrared data, characterized in that, include: Module 1: Acquire near-infrared fNIRS data of training users during the Stroop cognitive assessment task, and record the task execution results through a data storage device; Module 2: The coupling strength of HbO signal in brain regions at various frequency bands in the fNIRS data is obtained through dynamic Bayesian inference as a brain effect connectivity map. The data features of the brain effect connectivity map are extracted through the first convolutional neural network to obtain brain effect connectivity features. The data features of the fNIRS data are extracted through the second convolutional neural network to obtain spatiotemporal spectral features. Module 3: The brain effect connection feature, the spatiotemporal spectrum feature, and the data feature of the task execution result are concatenated to obtain the fusion feature. The classifier uses the fusion feature to obtain the probability that the training user belongs to AD, MCI, and healthy as the cognitive detection result. Based on the cognitive detection result and the training user's real cognitive detection label, a loss function is constructed to train the first convolutional neural network, the second convolutional neural network, and the classifier. Module 4: By inputting the brain effect connectivity map and fNIRS data of the user to be cognitively detected into the first convolutional neural network and the second convolutional neural network after training, respectively, the brain effect connectivity map and fNIRS data features of the user to be cognitively detected are obtained. After concatenation with the Stroop task execution features of the user to be cognitively detected, the data is input into the classifier after training to obtain the probability that the user to be cognitively detected belongs to AD, MCI and healthy, which serves as the cognitive detection support result for the user to be cognitively detected. The construction process of the brain effect connectivity map in Module 2 includes: The fNIRS data received by the functional near-infrared device is converted into blood oxygen data. The blood oxygen signal in the specified frequency band is obtained through filtering. The average blood oxygen concentration during the execution of the Stroop cognitive assessment task is calculated and wavelet transform is performed to obtain the signal change in the specified band. The average phase information of each frequency band is calculated using the phase angle function to derive the phase oscillation model between the two channels. The coupling strength of HbO signal in multiple brain regions at each frequency is calculated based on dynamic Bayesian inference to establish the brain effect connectivity map.

5. The cognitive detection support device based on the Stroop paradigm and near-infrared data as described in claim 4, characterized in that, The execution characteristics of this Stroop task include: Stroop task accuracy, error rate, and omission rate.

6. The cognitive detection support device based on the Stroop paradigm and near-infrared data as described in claim 4, characterized in that, Module 1 includes: presenting the Stroop cognitive assessment task to the training user and recording the training user's task completion status as the task execution result through a data storage device; and synchronously acquiring the training user's fNIRS data through a functional near-infrared device.

7. A server, characterized in that, Includes the cognitive detection support device based on the Stroop paradigm and near-infrared data as described in any one of claims 4-6.

8. A storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the cognitive detection support method based on the Stroop paradigm and near-infrared data as described in any one of claims 1-3.

Images