A neurocognitive function assessment method and device based on artificial intelligence

Abstract

The present invention provides a method and device for evaluating neurocognitive function based on artificial intelligence, and relates to the technical field of neurocognitive function evaluation. The method includes: collecting neuroimaging data, physiological signal data, and behavioral data; preprocessing the collected multiple neurological function cognitive evaluation data; extracting features from the preprocessed data to obtain corresponding feature data; constructing an evaluation model based on a capsule network combined with a graph convolutional neural network, training the model with a training set, and optimizing the model hyperparameters using the antlion algorithm; and evaluating the model through evaluation indicators. By collecting multiple data related to neurocognitive functions, the present invention constructs an evaluation model based on a capsule network combined with a graph convolutional neural network, and optimizes the model hyperparameters through the antlion algorithm, so as to better mine the hidden patterns in the neuroimaging data, realize a comprehensive evaluation of cognitive functions from multiple aspects, and improve the accuracy and applicability of neurocognitive function evaluation.

Description

Technical Field

[0001] The present invention relates to the technical field of neurocognitive function assessment, and in particular to a neurocognitive function assessment method and device based on artificial intelligence. Background Art

[0002] Neurocognitive function refers to the support and regulation of cognitive activities by the cerebral cortex and its neural network. It mainly includes the following aspects: attention and alertness, perception and memory, language function, executive function, spatial cognitive function, emotion and behavior regulation. These cognitive functions all rely on the normal operation of the cerebral cortex and its complex neural network.

[0003] When existing technologies use artificial intelligence to evaluate neurocognitive functions, the artificial intelligence system is unable to fully simulate and understand the mechanisms and processes of the brain's neurocognitive functions, and it is difficult to capture complex human cognitive activities. Artificial intelligence evaluation is more focused on certain specific cognitive function indicators, and it is difficult to conduct a comprehensive evaluation of cognitive functions from multiple aspects. Summary of the invention

[0004] The present invention provides a neurocognitive function assessment method and device based on artificial intelligence, which are used to solve the defect that it is difficult to comprehensively assess multiple aspects of cognitive function in the prior art.

[0005] In one aspect, the present invention provides a method for evaluating neurocognitive function based on artificial intelligence, comprising the following steps:

[0006] S1. Data collection: Collect data related to neurocognitive function, including neuroimaging data, physiological signal data, and behavioral data, and divide the data into training set, validation set, and test set;

[0007] S2. Data preprocessing: preprocess the collected data of various neurological function cognitive assessments to obtain corresponding preprocessed data;

[0008] S3, feature extraction: extract features from the preprocessed data to obtain corresponding feature data;

[0009] S4. Model construction and optimization: Build an evaluation model based on capsule network combined with graph convolutional neural network, use the training set to train the evaluation model of capsule network combined with graph convolutional neural network, and use the ant lion algorithm to optimize the model hyperparameters;

[0010] S5. Model evaluation and application: The accuracy, recall and F1 value are used as indicators to calculate and evaluate on the test set, and the model that meets the evaluation indicators is used to evaluate neurocognitive function.

[0011] According to an artificial intelligence-based neurocognitive function assessment method provided by the present invention, neuroimaging data include functional magnetic resonance imaging, positron emission tomography and electroencephalogram; physiological signal data include heart rate variability, skin conductivity and respiratory rate; behavioral data include reaction time, accuracy and eye tracking.

[0012] According to an artificial intelligence-based neurocognitive function assessment method provided by the present invention, the preprocessing method of functional magnetic resonance imaging data includes time correction, head motion correction, spatial standardization and denoising;

[0013] Preprocessing of positron emission computed tomography data includes attenuation correction, scatter correction, and image normalization;

[0014] The preprocessing methods of EEG data included baseline correction and filtering;

[0015] The preprocessing methods of heart rate variability data included noise removal and baseline correction;

[0016] The preprocessing methods of skin conductivity data included outlier removal and smoothing;

[0017] The preprocessing methods of respiratory rate data include removal of abnormal fluctuations and interpolation;

[0018] The preprocessing methods of reaction time data include outlier handling;

[0019] Preprocessing of accuracy data includes checking data integrity and merging or classifying;

[0020] The preprocessing methods of eye tracking data include data cleaning, coordinate transformation and smoothing.

[0021] According to an artificial intelligence-based neurocognitive function assessment method provided by the present invention, the feature extraction method of functional magnetic resonance imaging data is a functional connectivity analysis based on the Pearson correlation coefficient;

[0022] The feature extraction method of positron emission computed tomography data was standardized uptake value calculation;

[0023] The feature extraction method of EEG data is to convert the time domain signal into the frequency domain through fast Fourier transform and calculate the power spectral density of different frequency bands;

[0024] The feature extraction method of heart rate variability data is time domain analysis, that is, calculating the average normal heart beat interval;

[0025] The feature extraction method of skin conductivity data is to calculate the skin conductance level;

[0026] The feature extraction method of respiratory rate data is to calculate the average respiratory rate;

[0027] The method of extracting features from reaction time data is data normalization;

[0028] The feature extraction method of accuracy data is to calculate the overall accuracy;

[0029] The feature extraction of eye tracking data includes: gaze point distribution features, gaze time features and eye saccade features.

[0030] According to an artificial intelligence-based neurocognitive function assessment method provided by the present invention, the functional connectivity analysis based on the Pearson correlation coefficient is as follows:

[0031]

[0032] Among them, x and y represent the time series of two brain regions respectively. This formula calculates the ratio of the covariance of the two time series to the product of their respective standard deviations. When r is close to 1, it means that the activities of the two brain regions are highly correlated; when it is close to -1, it means negative correlation; when it is close to 0, it means almost no correlation;

[0033] Standardized uptake value calculation, the calculation formula is:

[0034] Standardized uptake value = radioactivity concentration in tissue / (injected radioactivity / patient weight);

[0035] The power spectral density calculation formula is:

[0036]

[0037] Where x[n] (n=0, 1, 2, ..., N-1) is the offline time signal, X[k] is the frequency domain value after transformation, K=0, 1, 2, ..., N-1 represents the frequency index, j is the imaginary unit, j 2 =-1, is a complex exponential term;

[0038] The average normal heart beat interval is calculated as follows:

[0039]

[0040] Among them, NN i is the i-th normal heartbeat interval, N is the total number of intervals, and the average heartbeat interval duration can be calculated by the above formula;

[0041] The formula for calculating skin conductance level is:

[0042]

[0043] Among them, SC iis the skin conductance value of the ith measurement, and N is the total number of measurements.

[0044] According to a neurocognitive function assessment method based on artificial intelligence provided by the present invention, the specific steps of capsule network construction are:

[0045] Capsule layer: A capsule is a group of neurons whose output is a vector. The specific formula is:

[0046] u j丨i =W ij u i

[0047] Among them, u i is the output vector of the lower-level capsule i, W ij is the weight matrix connecting the lower-level capsule i and the higher-level capsule j, u j丨i is the input vector of the predicted high-level capsule j.

[0048] Dynamic routing algorithm: Determine the connection strength between the lower-level capsules and the higher-level capsules. The formula is:

[0049]

[0050] Among them, c ij is the coupling coefficient from the lower-level capsule i to the higher-level capsule j, which is calculated by the initial logarithmic prior probability b ij Calculate the softmax. j is the output vector of the high-level capsule j, obtained by normalizing the weighted sum of the input vectors.

[0051] According to a neurocognitive function assessment method based on artificial intelligence provided by the present invention, the specific steps of constructing a graph convolutional neural network are:

[0052] Graph construction: Build a graph structure based on the anatomical or functional connections between brain regions, with nodes representing brain regions and edges representing the strength of connections between regions;

[0053] Graph convolutional layer: Its formula is:

[0054]

[0055] in, is the adjacency matrix with self-connection added, A is the original connectivity matrix, and I is the identity matrix. is the degree matrix, H l is the feature matrix of the lth layer, W l is the weight matrix of the lth layer, and σ is the activation function.

[0056] According to a neurocognitive function assessment method based on artificial intelligence provided by the present invention, a capsule network is combined with a graph convolutional neural network, and the specific steps are as follows:

[0057] Feature fusion: The features extracted by the capsule network and the features extracted by the graph convolutional neural network are fused. The formula is:

[0058] F combined =f(F capsule , F gcnn )

[0059] Among them, F capsule is the feature extracted by the capsule network, F gcnn is the feature extracted by the graph convolutional neural network, f is the fusion function, F combined It is the fusion feature;

[0060] Joint training: Use a common cross entropy loss function to train the fused model, the formula is:

[0061]

[0062] Among them, y i is the true label, is the probability predicted by the model.

[0063] According to a neurocognitive function assessment method based on artificial intelligence provided by the present invention, the specific steps of model training and ant lion algorithm optimization of model hyperparameters are:

[0064] Use the training set to train the capsule network combined with the graph convolutional neural network evaluation model;

[0065] The Ant Lion algorithm optimizes model hyperparameters. The specific steps are:

[0066] a: Initialization, defining the range of learning rate and initializing the ant lion population;

[0067] b: Define the fitness function and define the accuracy as the fitness function for evaluating the performance of hyperparameters;

[0068] c: Antlion position update

[0069] C1. Randomly generate ant positions

[0070] C2, update the ant position, the formula is:

[0071]

[0072] Among them, c t is the random step size control parameter that changes with the number of iterations t, c max and c minis the maximum and minimum value of the control parameter, T is the maximum number of iterations, ub j and lb j are the upper and lower limits of the jth hyperparameter, rand is a random number between [0, 1], and I is a random number. is the position of the ant in the jth dimension, is the position of the corresponding antlion in the jth dimension;

[0073] d: Calculate the accuracy. Use the hyperparameters corresponding to the updated ant positions to train the model and calculate the accuracy on the validation set.

[0074] e: Update the position of the ant lion. Update the position of the ant lion according to the fitness value of the ant. If the fitness value of an ant is better than the current ant lion, then update the position of the ant lion to the position of the ant;

[0075] f: Repeat steps c to e until the maximum number of iterations is reached;

[0076] g: Output the optimal hyperparameters, and finally output the hyperparameter combination corresponding to the antlion position with the optimal fitness value as the optimized model hyperparameters;

[0077] Retrain the model using the optimal hyperparameters.

[0078] On the other hand, the present invention also provides an artificial intelligence-based neurocognitive function assessment device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, an artificial intelligence-based neurocognitive function assessment method as claimed in any one of claims 1 to 9 is implemented.

[0079] The present invention provides an artificial intelligence-based neurocognitive function assessment method and device. According to the collected data related to various neurocognitive functions, the neurocognitive function assessment is performed by constructing a capsule network combined with a graph convolutional neural network assessment model, which solves the problem that it is difficult to conduct a comprehensive assessment of cognitive functions from multiple aspects in the prior art. The capsule network is good at capturing the spatial relationship between local features and can better model the complexity of neurocognitive functions. The graph convolutional neural network can use the topological structure information of brain imaging data to better learn the brain functional network characteristics. The ant lion algorithm can effectively search for the optimal hyperparameter combination of the model to improve the accuracy and generalization of the assessment. By constructing an assessment model based on a capsule network combined with a graph convolutional neural network and optimizing the model hyperparameters through the ant lion algorithm, the latest deep learning technology is integrated to better mine the hidden patterns in the neuroimaging data and improve the accuracy and applicability of the neurocognitive function assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0081] Figure 1 It is a flowchart of a method for evaluating neurocognitive function based on artificial intelligence provided by an embodiment of the present invention;

[0082] Figure 2 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.

[0083] Reference numerals:

[0084] 610, processor; 620, communication interface; 630, memory; 640, communication bus. DETAILED DESCRIPTION

[0085] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0086] Combine the following Figure 1-Figure 2 Specific embodiments of the present invention are described.

[0087] Figure 1 It is a flowchart of a neurocognitive function assessment method based on artificial intelligence provided by an embodiment of the present invention.

[0088] like Figure 1 As shown, an embodiment of the present invention provides a neurocognitive function assessment method based on artificial intelligence, the execution subject can be a neurocognitive function assessment device based on artificial intelligence, and the method mainly includes the following steps:

[0089] S1. Data collection: including neuroimaging data, physiological signal data and behavioral data. Neuroimaging data can provide information on the spatiotemporal distribution of brain activity; physiological signals reflect autonomic nervous activity and emotional state; behavioral data directly reflects the performance of cognitive function. Among them, neuroimaging data include functional magnetic resonance imaging, which collects blood oxygen level-dependent signals of the brain in different states by having individuals perform specific cognitive tasks or be in a resting state in a magnetic resonance scanner; positron emission tomography, which collects the distribution of tracers in the brain by injecting radioactive tracers into individuals to reflect the brain's metabolism or neurotransmitter activity; and electroencephalogram, which records the electrical activity of brain neurons by placing electrodes on the individual's head.

[0090] Physiological signal data include heart rate variability, which is calculated by using an electrocardiogram device to continuously record the heart beat intervals; skin conductivity, which is measured by placing electrodes on the fingers or palms to measure changes in skin conductance; and respiratory rate, which is monitored by using a respiratory belt or sensor to monitor the frequency and pattern of breathing.

[0091] Behavioral data include reaction time, which is to present stimuli through a computer program and record the time from the appearance of the stimulus to the individual's response; accuracy, which calculates the proportion of correct answers given by the individual in the task; and eye tracking, which uses an eye tracker to record the individual's eye movement trajectory and gaze point when performing the task.

[0092] In this embodiment, by comprehensively utilizing these multimodal data, an individual's cognitive function can be evaluated more comprehensively.

[0093] S2. Data preprocessing: preprocess the collected neurofunctional cognitive data to obtain preprocessed data. Among them, in neuroimaging data, the preprocessing methods of functional magnetic resonance imaging data include time correction: correct the time difference between different slices during the scanning process to align the time points of all slices; head movement correction: identify and correct image distortion caused by head movement, and align the images of all time points to a reference frame through transformations such as translation and rotation; spatial standardization: convert individual brain images into a standard template space so that the brain anatomical structures of different individuals correspond to the same coordinate position; denoising: including removing physiological noise and scanner noise.

[0094] The preprocessing methods of PET data include attenuation correction: used to correct the attenuation of photons in tissue; scatter correction: to correct the effect of scattered photons on the image; image normalization: to make the images between different scans comparable.

[0095] The preprocessing methods of EEG data include baseline correction: correcting the baseline by subtracting the mean potential; removing electrooculogram and electromyography artifacts: using independent component analysis method; filtering: using Butterworth filter to remove high-frequency noise and low-frequency drift, the formula is:

[0096]

[0097] Where H(s) is the transfer function, s is the complex frequency, and w c is the cutoff frequency and n is the order of the filter.

[0098] In physiological signal data, the preprocessing methods of heart rate variability data include noise removal: using low-pass filtering to remove high-frequency noise and high-pass filtering to remove low-frequency drift; baseline correction: correcting the baseline by subtracting the average heart rate or using polynomial fitting;

[0099] The preprocessing methods of skin conductivity data include removing outliers: identifying and removing abnormal conductivity values ​​by setting thresholds or based on statistical methods; smoothing: using moving average or median filtering to smooth the data;

[0100] The preprocessing methods of respiratory rate data include removing abnormal fluctuations: setting a reasonable range based on statistical methods or domain knowledge, and removing data points that exceed the range; interpolation processing: interpolation estimation for missing data points.

[0101] In the behavioral data, the preprocessing method of reaction time data includes outlier processing: by setting upper and lower thresholds, reaction time data that obviously deviates from the normal range are removed.

[0102] The preprocessing methods of accuracy data include checking data integrity: ensuring that the accuracy data of all tasks or experiments are fully recorded; merging or classification processing: merging or classifying the accuracy data according to different task conditions or groups.

[0103] The preprocessing methods of eye tracking data include data cleaning: removing invalid data points caused by blinking, head movement, etc.; smoothing: using the moving average method to reduce data jitter, the formula is:

[0104]

[0105] Among them, P new is the new eye position, P i It is n continuous eye movement position data; coordinate transformation: convert the original coordinates of the eye movement data into a coordinate system suitable for analysis.

[0106] In this embodiment, the above preprocessing steps can effectively improve data quality and accuracy, enhance data consistency and comparability, reduce data dimension and complexity, and lay the foundation for subsequent feature extraction.

[0107] S3. Feature extraction: Extract features from the preprocessed data to obtain corresponding feature data. The feature extraction method of functional magnetic resonance imaging data is functional connectivity analysis based on the Pearson correlation coefficient, and the formula is:

[0108]

[0109] Among them, x and y represent the time series of two brain regions respectively, and this formula calculates the ratio of the covariance of the two time series to the product of their respective standard deviations. When r is close to 1, it means that the activities of the two brain regions are highly correlated; when it is close to -1, it means negative correlation; when it is close to 0, it means almost no correlation.

[0110] The feature extraction method of PET data is: Standardized Uptake Value (SUV) calculation, the calculation formula is:

[0111] SUV = radioactivity concentration in tissue / (injected radioactivity / patient weight)

[0112] SUV is used to quantify the extent of radiotracer uptake in tissues and reflects the metabolic activity of the tissue.

[0113] The feature extraction method of EEG data is to convert the time domain signal into the frequency domain through fast Fourier transform and calculate the power spectrum density of different frequency bands. The formula is:

[0114]

[0115] Where x[n] (n=0, 1, 2, ..., N-1) is the offline time signal, X[k] is the frequency domain value after transformation, K=0, 1, 2, ..., N-1 represents the frequency index, j is the imaginary unit, j 2 =-1, is a complex exponential term.

[0116] The feature extraction method of heart rate variability data is time domain analysis, that is, calculating the average normal heartbeat interval, and the formula is:

[0117]

[0118] Among them, NN i is the i-th normal heartbeat interval, N is the total number of intervals, and the average heartbeat interval duration can be calculated using the above formula.

[0119] The feature extraction method of skin conductivity data is to calculate the skin conductance level, and the formula is:

[0120]

[0121] Among them, SC i is the skin conductance value of the ith measurement, and N is the total number of measurements.

[0122] The feature extraction method of respiratory rate data is to calculate the average respiratory rate, and the formula is:

[0123]

[0124] The average number of breaths per minute is calculated by counting the number of breaths taken in a given period of time and dividing that time by the breaths taken.

[0125] The method of extracting reaction time data features is data standardization, and the formula is:

[0126]

[0127] Where X is the raw reaction time value, μ is the mean reaction time, and μ is the standard deviation of the reaction time.

[0128] The feature extraction method of accuracy data is to calculate the overall accuracy, and the formula is:

[0129]

[0130] By calculating the overall accuracy, the proportion of correct responses in the entire task can be directly calculated, which simply and intuitively reflects the overall performance level.

[0131] The feature extraction of eye tracking data includes: gaze point distribution feature, gaze time feature and saccade feature. Among them, the gaze point distribution feature describes the distribution by calculating the number or time proportion of gaze points in different areas; the gaze time feature includes the average gaze time, and its formula is average gaze time = total gaze time / number of gazes; the saccade feature includes the average saccade amplitude, and its formula is average saccade amplitude = total saccade amplitude / number of saccades.

[0132] S4. Model construction and optimization: Construct an evaluation model based on capsule network combined with graph convolutional neural network. The specific steps are as follows:

[0133] S41. Build capsule network

[0134] Capsule layer: A capsule is a group of neurons whose output is a vector. The specific formula is:

[0135] u j丨i =W ij u i

[0136] Among them, u i is the output vector of the lower-level capsule i, W ij is the weight matrix connecting the lower-level capsule i and the higher-level capsule j, u j丨i is the input vector of the predicted high-level capsule j.

[0137] Dynamic routing algorithm: Determine the connection strength between the lower-level capsules and the higher-level capsules. The formula is:

[0138]

[0139] Among them, c ij is the coupling coefficient from the lower-level capsule i to the higher-level capsule j, which is calculated by the initial logarithmic prior probability b ij Calculate the softmax. j is the output vector of the high-level capsule j, obtained by normalizing the weighted sum of the input vectors.

[0140] S42. Constructing graph convolutional neural network

[0141] Graph construction: A graph structure is constructed based on the anatomical or functional connections between brain regions, with nodes representing brain regions and edges representing the strength of connections between regions.

[0142] Graph convolutional layer: Its formula is:

[0143]

[0144] in, is the adjacency matrix with self-connection added, A is the original connectivity matrix, and I is the identity matrix. is the degree matrix, H l is the feature matrix of the lth layer, W l is the weight matrix of the lth layer, and σ is the activation function.

[0145] S43. Combining capsule network with graph convolutional neural network

[0146] S431: Feature fusion: Fusion of features extracted by capsule network and graph convolutional neural network. The formula is:

[0147] F combined =f(F capsule , F gcnn )

[0148] Among them, F capsule is the feature extracted by the capsule network, F gcnn is the feature extracted by the graph convolutional neural network, f is the fusion function, F combined It is the fusion feature.

[0149] S432: Joint training: Use a common cross entropy loss function to train the fused model. The formula is:

[0150]

[0151] Among them, y i is the true label, is the probability predicted by the model.

[0152] S44, model training and antlion algorithm optimization

[0153] S441: Use the training set to train the capsule network combined with the graph convolutional neural network evaluation model

[0154] S442: Antlion Algorithm Optimizes Model Hyperparameters

[0155] a: Initialization, defining the range of learning rate and initializing the ant lion population;

[0156] b: Define the fitness function and define the accuracy as the fitness function for evaluating the performance of hyperparameters;

[0157] c: Antlion position update

[0158] C1. Randomly generate ant positions

[0159] C2, update the ant position, the formula is:

[0160]

[0161] Among them, c t is the random step size control parameter that changes with the number of iterations t, c max and c min is the maximum and minimum value of the control parameter, T is the maximum number of iterations, ub j and lb j are the upper and lower limits of the jth hyperparameter, rand is a random number between [0, 1], and I is a random number. is the position of the ant in the jth dimension, is the position of the corresponding antlion in the jth dimension.

[0162] d: Calculate the accuracy. Use the hyperparameters corresponding to the updated ant positions to train the model and calculate the accuracy on the validation set.

[0163] e: Update the position of the ant lion. Update the position of the ant lion according to the fitness value of the ant. If the fitness value of an ant is better than the current ant lion, then update the position of the ant lion to the position of the ant.

[0164] f: Repeat steps c to e until the predetermined maximum number of iterations is reached.

[0165] g: Output the optimal hyperparameters, and finally output the hyperparameter combination corresponding to the antlion position with the optimal fitness value as the optimized model hyperparameters.

[0166] S443: Retrain the model using optimal hyperparameters.

[0167] S5. Model evaluation and application

[0168] The accuracy, recall and F1 value are used as indicators to calculate the evaluation on the test set, and the model that meets the evaluation indicators is used to evaluate the neurocognitive function.

[0169] In this embodiment, based on the collected data related to various neurocognitive functions, by constructing a capsule network combined with a graph convolutional neural network evaluation model for neurocognitive function evaluation, the accuracy and generalization performance can be improved, and a comprehensive evaluation of cognitive functions from multiple aspects can be achieved, thereby improving the accuracy and applicability of neurocognitive function evaluation. The capsule network can capture the spatial relationships and hierarchical features in the input data, which is beneficial to improving the model's ability to recognize complex neurocognitive functions. The graph convolutional neural network can use brain structure information to better model the spatial characteristics of neural activity. The combination of the two network structures can make full use of the spatial, hierarchical and structural information of the input data, thereby improving the model's evaluation accuracy and generalization performance for neurocognitive functions.

[0170] By optimizing the hyperparameters of the capsule network combined with the graph convolutional neural network evaluation model using the antlion algorithm, we can effectively search for the hyperparameters and structure of the network model and find a better configuration, which can further improve the accuracy and generalization performance of the model in the task of neurocognitive function evaluation. This can significantly improve the performance, robustness and interpretability of the model, speed up the training and deployment of the model, and bring positive impacts to the evaluation of neurocognitive function.

[0171] Based on the same general inventive concept, the present invention also protects a neurocognitive function assessment device based on artificial intelligence. The neurocognitive function assessment device based on artificial intelligence provided by the present invention is described below. The neurocognitive function assessment device based on artificial intelligence described below and the neurocognitive function assessment method based on artificial intelligence described above can be referenced to each other.

[0172] The neurocognitive function assessment device may include: a processor 610, a communications interface 620, a memory 630, and a communication bus 640, wherein the processor 610, the communications interface 620, and the memory 630 communicate with each other via the communication bus 640. The processor 610 may call the logic instructions in the memory 630 to execute a neurocognitive function assessment method based on artificial intelligence.

[0173] In addition, the logic instructions in the above-mentioned memory 630 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0174] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute an artificial intelligence-based neurocognitive function assessment method provided by the above methods.

[0175] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute an artificial intelligence-based neurocognitive function assessment method provided by the above methods.

[0176] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0177] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0178] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A neurocognitive function assessment method based on artificial intelligence, characterized in that: The steps include: S1. Data collection: Collect data related to neurocognitive function, including neuroimaging data, physiological signal data, and behavioral data, and divide the data into training set, validation set, and test set; S2. Data preprocessing: preprocess the collected data of various neurological function cognitive assessments to obtain corresponding preprocessed data; S3, feature extraction: extract features from the preprocessed data to obtain corresponding feature data; S4. Model construction and optimization: Build an evaluation model based on capsule network combined with graph convolutional neural network, use the training set to train the evaluation model of capsule network combined with graph convolutional neural network, and use the ant lion algorithm to optimize the model hyperparameters; The capsule network is combined with the graph convolutional neural network. The specific steps are: Feature fusion: The features extracted by the capsule network and the features extracted by the graph convolutional neural network are fused. The formula is: F combined =f(F capsule ,F gcnn ) Among them, F capsule is the feature extracted by the capsule network, F gcnn is the feature extracted by the graph convolutional neural network, f is the fusion function, F combined It is the fusion feature; Joint training: Use a common cross entropy loss function to train the fused model, the formula is: Among them, y i is the true label, is the probability predicted by the model; The specific steps of model training and Ant Lion algorithm optimization of model hyperparameters are: Use the training set to train the capsule network combined with the graph convolutional neural network evaluation model; The Ant Lion algorithm optimizes model hyperparameters. The specific steps are: a: Initialization, defining the range of learning rate and initializing the ant lion population; b: Define the fitness function and define the accuracy as the fitness function for evaluating the performance of hyperparameters; c: Antlion position update C1. Randomly generate ant positions C2, update the ant position, the formula is: Among them, c t is the random step size control parameter that changes with the number of iterations t, c max and c min is the maximum and minimum value of the control parameter, T is the maximum number of iterations, ub j and lb j are the upper and lower limits of the jth hyperparameter, rand is a random number between [0, 1], and I is a random number. is the position of the ant in the jth dimension, is the position of the corresponding antlion in the jth dimension; d: Calculate the accuracy. Use the hyperparameters corresponding to the updated ant positions to train the model and calculate the accuracy on the validation set. e: Update the position of the ant lion. Update the position of the ant lion according to the fitness value of the ant. If the fitness value of an ant is better than the current ant lion, then update the position of the ant lion to the position of the ant. f: Repeat steps c to e until the maximum number of iterations is reached; g: Output the optimal hyperparameters, and finally output the hyperparameter combination corresponding to the antlion position with the optimal fitness value as the optimized model hyperparameters; Retrain the model using optimal hyperparameters; S5. Model evaluation and application: The accuracy, recall and F1 value are used as indicators to calculate and evaluate on the test set, and the model that meets the evaluation indicators is used to evaluate neurocognitive function.

2. The method for evaluating neurocognitive function based on artificial intelligence according to claim 1, characterized in that: Neuroimaging data included functional magnetic resonance imaging, positron emission tomography, and electroencephalography; physiological signal data included heart rate variability, skin conductivity, and respiration rate; and behavioral data included reaction time, accuracy, and eye tracking.

3. The method for evaluating neurocognitive function based on artificial intelligence according to claim 2, characterized in that: Preprocessing of fMRI data included temporal correction, head motion correction, spatial normalization, and denoising; Preprocessing of positron emission computed tomography data includes attenuation correction, scatter correction, and image normalization; The preprocessing methods of EEG data included baseline correction and filtering; The preprocessing methods of heart rate variability data included noise removal and baseline correction; The preprocessing methods of skin conductivity data included outlier removal and smoothing; The preprocessing methods of respiratory rate data include removal of abnormal fluctuations and interpolation; The preprocessing methods of reaction time data include outlier handling; Preprocessing of accuracy data includes checking data integrity and merging or classifying; The preprocessing methods of eye tracking data include data cleaning, coordinate transformation and smoothing.

4. The method for evaluating neurocognitive function based on artificial intelligence according to claim 3, characterized in that: Features of functional magnetic resonance imaging data were extracted using functional connectivity analysis based on Pearson correlation coefficient; The feature extraction method of positron emission computed tomography data was standardized uptake value calculation; The feature extraction method of EEG data is to convert the time domain signal into the frequency domain through fast Fourier transform and calculate the power spectral density of different frequency bands; The feature extraction method of heart rate variability data is time domain analysis, that is, calculating the average normal heart beat interval; The feature extraction method of skin conductivity data is to calculate the skin conductance level; The feature extraction method of respiratory rate data is to calculate the average respiratory rate; The method of extracting features from reaction time data is data normalization; The feature extraction method of accuracy data is to calculate the overall accuracy; The feature extraction of eye tracking data includes: gaze point distribution features, gaze time features and eye saccade features.

5. The method for evaluating neurocognitive function based on artificial intelligence according to claim 4, characterized in that: Functional connectivity analysis based on Pearson correlation coefficient, the formula is: Among them, x and y represent the time series of two brain regions respectively. This formula calculates the ratio of the covariance of the two time series to the product of their respective standard deviations. When r is close to 1, it means that the activities of the two brain regions are highly correlated; when it is close to -1, it means negative correlation; when it is close to 0, it means almost no correlation; Standardized uptake value calculation, the calculation formula is: Standardized uptake value = radioactivity concentration in tissue / (injected radioactivity / patient weight); The power spectral density calculation formula is: Where x[n] (n=0, 1, 2, ..., N-1) is the offline time signal, X[k] is the frequency domain value after transformation, K=0, 1, 2, ..., N-1 represents the frequency index, j is the imaginary unit, j 2 =-1, is a complex exponential term; The average normal heart beat interval is calculated as follows: Among them, NN i is the i-th normal heartbeat interval, N is the total number of intervals, and the average heartbeat interval duration can be calculated by the above formula; The formula for calculating skin conductance level is: Among them, SC i is the skin conductance value of the ith measurement, and N is the total number of measurements.

6. The method for evaluating neurocognitive function based on artificial intelligence according to claim 1, characterized in that: The specific steps of capsule network construction are: Capsule layer: A capsule is a group of neurons whose output is a vector. The specific formula is: u j丨i =W ij u i Among them, u i is the output vector of the lower-level capsule i, W ij is the weight matrix connecting the lower-level capsule i and the higher-level capsule j, u j丨i is the input vector of the predicted high-level capsule j; Dynamic routing algorithm: Determine the connection strength between the lower-level capsules and the higher-level capsules. The formula is: Among them, c ij is the coupling coefficient from the lower-level capsule i to the higher-level capsule j, which is calculated by the initial logarithmic prior probability b ij Perform softmax calculation to get, v j is the output vector of the high-level capsule j, obtained by normalizing the weighted sum of the input vectors.

7. The method for evaluating neurocognitive function based on artificial intelligence according to claim 1, characterized in that: The specific steps of constructing a graph convolutional neural network are: Graph construction: Build a graph structure based on the anatomical or functional connections between brain regions, with nodes representing brain regions and edges representing the strength of connections between regions; Graph convolutional layer: Its formula is: in, is the adjacency matrix with self-connection added, A is the original connection matrix, I is the identity matrix, is the degree matrix, H l is the feature matrix of the lth layer, W l is the weight matrix of the lth layer, and σ is the activation function.

8. A neurocognitive function assessment device based on artificial intelligence, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, an artificial intelligence-based neurocognitive function assessment method as described in any one of claims 1 to 7 is implemented.

Images