Model tuning method and device based on neural image and cognitive behavior

Abstract

The invention relates to the technical field of computers, and discloses a model tuning method and device based on neural images and cognitive behaviors, and the method comprises the steps: obtaining the scanned neural image data of an experimental subject in the process of executing the cognitive behaviors; performing structural feature extraction and functional feature extraction on the neural image data; according to the method, structural features and functional features are obtained, the structural features and the functional features are converted into parameters with physiological significance in a pre-constructed neurodynamic model, parameter optimization is conducted on the neurodynamic model, and the pre-constructed neurodynamic model and an experimental subject reflect the same change trend. According to the method, the neural image data are analyzed, the obtained structural features and functional features are converted into parameters with physiological significance in the neurodynamic model, parameter tuning is carried out on the neurodynamic model, the performance of the neurodynamic model is improved, and the parameter tuning process of the neurodynamic model has physiological interpretability.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a model tuning method and device based on neuroimaging and cognitive behavior. Background Art

[0002] At present, in the field of artificial intelligence, one of the important research methods is to imitate the structural characteristics of the biological nervous system, simplify and construct corresponding mathematical models to reproduce the intelligent behavior of organisms, which has been successful in the fields of pattern recognition and machine vision.

[0003] In the existing technology, the construction of artificial intelligence neural networks represented by machine learning and deep learning usually adopts a general model framework. However, the training and tuning of subsequent networks currently rely on large-scale data-driven and parameter value exploration in high-dimensional parameter space, and most of them blindly search for the optimal result in the parameter space, making the model lack of physiological interpretability. Summary of the invention

[0004] In view of this, the present invention provides a model tuning method and device based on neuroimaging and cognitive behavior to solve the problem in the prior art that parameter tuning relies on large-scale data driving and parameter values ​​in high-dimensional parameter space, resulting in the lack of physiological interpretability of the model.

[0005] In a first aspect, the present invention provides a model tuning method based on neuroimaging and cognitive behavior, the method comprising:

[0006] Obtaining neuroimaging data of scanned experimental subjects while they are performing cognitive behaviors;

[0007] Extract structural and functional features from neuroimaging data;

[0008] The extracted structural and functional features are converted into physiologically meaningful parameters in a pre-built neural dynamics model, and the parameters of the neural dynamics model are optimized. The pre-built neural dynamics model and the experimental object show the same change trend.

[0009] The present invention scans neural images during the cognitive process of experimental subjects, analyzes the scanned neural image data, extracts structural features and functional features, performs parameter conversion on the structural features and functional features, pre-constructs a neural dynamics model, converts the structural features and functional features into parameters with physiological significance in the neural dynamics model, so as to achieve parameter tuning of the neural dynamics model, and at the same time improves the performance of the neural dynamics model and makes the parameter tuning process of the neural dynamics model physiologically interpretable.

[0010] In an optional embodiment, the method further includes:

[0011] Obtain historical neuroimaging experimental results and extract dynamic characteristics from them. The dynamic characteristics are used to describe the bioelectric activity characteristics generated by neuronal cell bodies and synaptic connections between neurons.

[0012] The extracted dynamic characteristics are represented by differential equations and a neural dynamics model is constructed.

[0013] The present invention describes the dynamic characteristics of bioelectric activities generated by neuronal cell bodies and synaptic connections between neurons in the form of differential equations, and constructs a neurodynamic model through neurodynamic modeling to facilitate analysis of neuroimaging data of experimental subjects.

[0014] In an optional embodiment, when the cognitive behavior is motion perception, the extracted dynamic characteristics are expressed using differential equations to construct a neural dynamic model, including:

[0015] According to the extracted dynamic features, the dorsal visual pathway of the biological brain that performs motion perception tasks and several key brain regions corresponding to the dorsal visual pathway are determined, and the neuron models of each key brain region are constructed respectively;

[0016] Based on a set of differential equations, a neural dynamics model is constructed according to the neuron models of each key brain area. The neural dynamics model is used to simulate the dorsal visual pathway of the biological brain.

[0017] The present invention achieves brain region division by determining the key brain areas corresponding to the dorsal visual pathway of the biological brain that performs motion perception tasks, so as to construct neuron models of each key brain area respectively, and then establishes a neural dynamics model based on a set of differential equations to complete the simulation of the dorsal visual pathway of the biological brain, so as to analyze the neural dynamics model using neural imaging data.

[0018] In an optional embodiment, when the cognitive behavior is motion perception, after constructing the neural dynamics model, the method further includes:

[0019] Collect random point motion videos and input the random point motion videos into the neural dynamics model for performance testing;

[0020] The motion direction output by the neural dynamics model is compared with the actual motion direction to measure the sensitivity of the neural dynamics model.

[0021] The present invention inputs the collected random point motion video into the neural dynamics model to perform a sensitive performance test on the neural dynamics model, so as to optimize the neural dynamics model, enhance the performance of the neural dynamics model, and enhance the physiological interpretability of the neural dynamics model.

[0022] In an optional embodiment, when the cognitive behavior is motion perception, structural feature extraction and functional feature extraction are performed on the neuroimaging data, including:

[0023] Brain area segmentation is performed based on the brain structure partition map and brain function partition map;

[0024] Extracting structural features of key brain regions from neuroimaging data based on brain structural partitioning maps, including gray matter parameters, cortical parameters, and white matter parameters;

[0025] Functional characteristics of key brain regions are extracted from neuroimaging data based on the brain functional partitioning map, and the functional characteristics include resting-state functional connectivity data between key brain regions.

[0026] The present invention performs brain region segmentation based on the brain structure partition map and the brain function partition map, so as to extract structural features and functional features according to the brain structure partition map and the brain function partition map respectively, thereby realizing feature extraction of neuroimaging data.

[0027] In an optional embodiment, when the cognitive behavior is motion perception, the extracted structural features and functional features are converted into physiologically meaningful parameters in a pre-built neural dynamics model, and the parameters of the neural dynamics model are tuned, including:

[0028] The extracted structural and functional features are correlated with the previously acquired behavioral quantitative indicators to screen out the features that are significantly correlated with the behavioral quantitative indicators;

[0029] Convert features that are significantly correlated with behavioral quantitative indicators into physiologically meaningful parameters in the neural dynamics model, including synaptic connection parameters and synaptic connection probabilities;

[0030] Parameter adjustment is performed on synaptic connection parameters and synaptic connection probabilities in the neural dynamics model.

[0031] The present invention performs correlation analysis between the extracted structural features and functional features and the pre-acquired behavioral quantitative indicators when the cognitive behavior is motor perception, so as to screen the features and obtain the features that are significantly correlated with the behavioral quantitative indicators, perform parameter conversion on the features that are significantly correlated with the behavioral quantitative indicators, and perform parameter adjustment on the parameters with physiological significance in the neurodynamic model, thereby realizing parameter tuning of the neurodynamic model and making the parameter tuning process have physiological significance, which reflects the high efficiency and interpretability in the parameter tuning process.

[0032] In an optional embodiment, the method further includes:

[0033] When it is detected that the performance of the neural dynamics model has not changed significantly, the parameter adjustment of the neural dynamics model is stopped.

[0034] The present invention ensures that the performance of the neural dynamics model is at a high level by not adjusting the parameters of the neural dynamics model when the performance of the neural dynamics model does not change significantly.

[0035] In a second aspect, the present invention provides a model tuning device based on neuroimaging and cognitive behavior, the device comprising:

[0036] An acquisition module, used to acquire the neuroimaging data of the scanned experimental subject during the cognitive behavior;

[0037] A feature extraction module is used to extract structural features and functional features from neuroimaging data;

[0038] The model tuning module is used to convert the extracted structural features and functional features into physiologically meaningful parameters in the pre-built neural dynamics model, and to tune the parameters of the neural dynamics model. The pre-built neural dynamics model and the experimental object show the same change trend.

[0039] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the model tuning method based on neuroimaging and cognitive behavior according to the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0040] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the model tuning method based on neuroimaging and cognitive behavior according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 is a flowchart of a model tuning method based on neuroimaging and cognitive behavior according to an embodiment of the present invention;

[0043] Figure 2is a schematic diagram of key brain regions corresponding to the dorsal visual pathway according to an embodiment of the present invention;

[0044] Figure 3 is an overall framework diagram of a neural dynamics model according to an embodiment of the present invention;

[0045] Figure 4a is a schematic diagram for characterizing the relationship between the V1-MT white matter connection brain area and its average FA value and behavioral performance according to an embodiment of the present invention;

[0046] Figure 4b is a schematic diagram for characterizing the relationship between MT-LIP inter-white matter connected brain regions and their white matter volumes and behavioral performance according to an embodiment of the present invention;

[0047] Figure 4c is a schematic diagram for characterizing the relationship between the facultative lobe AACIC brain region and its MT-AAIC resting functional connectivity and behavioral performance according to an embodiment of the present invention;

[0048] Figure 4d is a schematic diagram for characterizing the relationship between the prefrontal suborbital brain region and its LIP-suborbital brain region and behavioral performance according to an embodiment of the present invention;

[0049] Figure 5a is a schematic diagram of changes in the mean distribution of the number of prominent connections between V1 and MT and its corresponding decision time and model performance according to an embodiment of the present invention;

[0050] Figure 5b is a schematic diagram of changes in the mean distribution of the number of prominent connections between MT-LIPs and its corresponding decision time and model performance according to an embodiment of the present invention;

[0051] Figure 5c is a schematic diagram of changes in excitatory synaptic conductivity between MT-LIP and its corresponding decision time and model performance according to an embodiment of the present invention;

[0052] Figure 5d is a schematic diagram of changes in the conductance of LIP excitatory synapses and their corresponding decision time and model performance according to an embodiment of the present invention;

[0053] Figure 6 is a structural block diagram of a model tuning device based on neuroimaging and cognitive behavior according to an embodiment of the present invention;

[0054] Figure 7 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0056] At present, one of the important research methods in the field of artificial intelligence is to imitate the structural characteristics of biological nervous systems, simplify and construct corresponding mathematical models to reproduce the intelligent behavior of organisms. CNN (Convolutional Neural Network), based on the receptive field model of animal vision and parallel distributed processing technology, such as AlexNet, GoogLeNet, ResNet, etc., has reached or even exceeded the average level in many fields such as target detection and face recognition. These DCNN (Deep Convolutional Neural Network) are not only used in image recognition and other fields, but also gradually extended from 2D image data processing to 2+1D video data processing, and have been widely used in moving target detection, action recognition and other fields. Despite this, DCNN is also plagued by defects such as large training sets, poor interpretability, and susceptibility to adversarial sample attacks. Therefore, drawing on the knowledge of neuroscience to develop brain-like algorithms to overcome these defects is an important direction of artificial intelligence research.

[0057] The current construction of artificial intelligence neural networks represented by machine learning and deep learning usually adopts a general model framework. Although these frameworks are inspired by the results of neuroscience research, such as the convolution process, which comes from neuroscience's understanding of the functions and structural characteristics of the brain's primary visual cortex in processing retinal input information, the subsequent network training and tuning all rely on data-driven objective function optimization methods, that is, using a large amount of training data to mathematically find the optimal weights between network connections for the optimization goal, thereby improving model performance.

[0058] The current data-driven model training method makes the model lack of interpretability. For example, weight tuning is mostly blindly searching for the optimal result in the parameter space, which lacks physiological interpretability. At the same time, the distribution of data will have a significant impact on the final training results. The convergence direction of model optimization will be affected by the data distribution, causing problems such as overfitting. Many training methods do not update and iterate the model in the right direction, but often fall into the misunderstanding of local optimal solutions, which greatly increases the difficulty of subsequent optimization of the model and the lack of generalization of the model when used.

[0059] Compared with biological intelligence, the model construction and training methods based solely on data drive make the entire process and results lack interpretability. By simulating the composition of biological intelligence and using the method of neural dynamics modeling to describe the dynamic characteristics of neurons and synaptic connections in the form of a set of differential equations, artificial neural networks with simulated biological functions can also be realized, and the problem of lack of interpretability of the model can be effectively solved.

[0060] This type of modeling based on neurodynamics has undergone simplification and dimensionality reduction of the biological model, and the values ​​of the neuron parameters are usually determined by referring to physiological experimental data. Due to the abstraction of the model and the simplification of the biological brain, as well as the multi-level structure that the modeling focuses on, it is difficult to directly correspond the measured data in actual physiological experiments, especially the more macroscopic brain area connectivity characteristics, to the model parameters. In the field of dynamic modeling, there are also few works that use the performance of human experts in behavioral experiments and their physiological characteristics to guide model construction or interpret important model parameters. And due to the characteristics of the dynamic system, it is impossible to simply use data-driven methods to numerically calculate and solve the optimal network connection characteristics and other parameters. The tuning and training of the dynamic system has become a difficulty that needs to be solved.

[0061] At present, a series of studies have been devoted to exploring the relationship between biological subject behavior and biological brain structural physiological characteristics. The biological behaviors focused on in these studies include various cognitive functions, such as working memory, language acquisition, and social functions such as the size of social networks. There are also studies focusing on the differences in physiological characteristics between healthy and diseased organisms, such as the correlation analysis between neurodegenerative diseases and physiological indicators of white matter fiber bundles. Neuroimaging technology provides technical support for quantifying biological brain physiological characteristics for such studies. The physiological characteristics used in these studies include: gray matter volume and cortical thickness calculated in structural image data; quantitative indicators of white matter areas in diffusion image data, such as FA value (Fractional Anisotropy) and MD value (Mean Diffusivity). Resting-state functional connectivity in functional image data, that is, measuring the correlation between blood oxygen level signals between brain regions in a spontaneously active state, is used as a functional connectivity indicator with endogenous properties and is widely used in the study of the correlation between biological brain function and behavior. These studies have revealed, to a certain extent, the correlation between brain structural and functional characteristics and biological behavioral performance, and have been applied to study the mechanisms of disease or the generation of intelligent behavior. However, the results of such studies have not been fully applied to the construction or improvement of artificial intelligence models.

[0062] In the related art, the existing artificial neural network modeling methods rely on large-scale data driving and parameter value exploration in high-dimensional parameter space to tune the model performance. The high-dimensional characteristics of the parameter space make it easy for the parameter tuning to fall into the local optimum rather than obtain the global optimal solution, and the tuning process also lacks interpretability. The present invention uses the specific structural and functional characteristics obtained from the "expert subject" data to perform parameter tuning on the feature-related parameters on the neural dynamics model, so that the model performance is improved while reducing the dimension of the parameter space, that is, reducing the difficulty of parameter tuning, and at the same time, the tuning process has physiological significance, that is, interpretability.

[0063] The embodiment of the present invention provides a model tuning method based on neuroimaging and cognitive behavior, a neurodynamic model that conforms to the physiological structure characteristics of the biological brain, and through neuroimaging and behavioral experiments, obtains the differences in structural and functional characteristics in specific brain regions between "expert subjects" with excellent performance and "ordinary subjects" with average performance when completing the same task, and based on the discovered characteristics, adjusts the corresponding model parameters in a targeted manner to achieve performance improvement of the artificial intelligence model. This method not only retains the interpretability of the neurodynamic system based on the biological brain structure modeling, but also provides clear goals and methods for tuning the model performance, thereby solving the problem of lack of interpretability of the machine learning model training process and results and lack of tuning basis for the neurodynamic model.

[0064] Among them, neurodynamics is a discipline theory that regards neural networks as a nonlinear dynamic system, studies the nature and mechanism of the change of the state of the nervous system over time, and reveals the operating laws of the nervous system. In mathematics, dynamic systems are usually described as a system of differential equations or difference equations of the system state with respect to time. Specifically, neurodynamics is based on the physiological and anatomical basis of the nervous system, and establishes differential equations related to the state of the nervous system. It describes the changes in the probability of ion channel opening and neuronal membrane potential at the microscopic level, and describes the spatiotemporal pattern of neuronal group activity generated by synaptic interactions between neurons at the mesoscopic level. It describes the spatiotemporal pattern of signal transmission and overall activity between various parts of the brain and the cognitive functions they support at the macroscopic level.

[0065] The neurodynamic model is an artificial intelligence neural network model based on a set of differential equations, which is formed by representing individual neuron cells and the synaptic connections between neuron cells in the form of differential equations and is used to complete a specific task.

[0066] Neuroimaging is an imaging method that uses various physical signal feature acquisition methods (such as magnetic resonance imaging, magnetoencephalography, electroencephalography, optical microscopy, scanning electron microscopy, etc.) to collect brain neural activities at different scales.

[0067] According to an embodiment of the present invention, an embodiment of a model tuning method based on neuroimaging and cognitive behavior is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0068] In this embodiment, a model tuning method based on neuroimaging and cognitive behavior is provided. Figure 1 is a flow chart of a model tuning method based on neuroimaging and cognitive behavior according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0069] Step S101, obtaining the scanned neuroimaging data of the experimental subject in the process of performing cognitive behavior.

[0070] In the embodiment of the present invention, while the experimental subject is performing cognitive behavior, behavioral experiments and neuroimaging scanning experiments are conducted to obtain neuroimaging data to acquire brain structure and brain function data information of the experimental subject.

[0071] When collecting neuroimaging data, the design of the cognitive behavioral tasks performed by the experimental subjects can be appropriately adjusted according to the different characteristics of the neuroimaging modality acquisition, so as to ensure that the data acquisition method is consistent with the neuroimaging data acquisition method, while being able to observe the dynamic activity data of the relevant brain areas corresponding to the model or the static structure between the relevant areas of the biological brain.

[0072] Step S102: extracting structural features and functional features from the neural imaging data.

[0073] In an embodiment of the present invention, the neuroimaging data is analyzed, including extracting the structural features and functional features of the individual brain. Taking the neurodynamic model as a motor perception neurodynamic model as an example, the neuroimaging data is analyzed, and the behavioral indicators are used as a reference to extract the imaging parameters and features related to the good or bad behavioral indicators in the neuroimaging data, including but not limited to structural connection features, functional connection features, etc.

[0074] Step S103, converting the extracted structural features and functional features into parameters with physiological significance in a pre-built neural dynamics model, and performing parameter tuning on the neural dynamics model.

[0075] In the embodiment of the present invention, the pre-built neural dynamics model and the experimental subject present similar behavior curve characteristics, that is, the cognitive behavior completed by the experimental subject and the cognitive behavior in the pre-built neural dynamics model show the same change trend. The extracted structural features and functional features are converted into parameters with physiological significance in the pre-built neural dynamics model to explore the value of the parameters, thereby optimizing the model performance.

[0076] The model tuning method based on neuroimaging and cognitive behavior provided in this embodiment scans neuroimaging while the experimental subject is performing cognitive processing, analyzes the scanned neuroimaging data, extracts structural features and functional features, performs parameter conversion on the structural features and functional features, pre-builds a neurodynamic model, converts the structural features and functional features into physiologically meaningful parameters in the neurodynamic model, so as to achieve parameter tuning of the neurodynamic model, and make the parameter tuning process of the neurodynamic model physiologically interpretable while improving the performance of the neurodynamic model.

[0077] In this embodiment, a model tuning method based on neuroimaging and cognitive behavior is provided, and the process includes the following steps:

[0078] Step S201, obtaining historical neuroimaging experimental results, and extracting dynamic characteristics from the historical neuroimaging experimental results.

[0079] Step S202, using differential equations to represent the extracted dynamic characteristics and constructing a neural dynamics model.

[0080] In an embodiment of the present invention, historical neuroimaging experimental results can be obtained from historical experiments or from historical documents. The basic components of the neural dynamics model are constructed by describing the dynamic characteristics of the bioelectric activity generated by the neuronal cell bodies and the synaptic connections between neurons through a group of differential equations. The basic components include but are not limited to neuron models and neuron synapse models, and their main parameters include but are not limited to the number of neuron models and the number of neuron synapse models.

[0081] Based on the results of historical neuroimaging experiments, an artificial neural network based on neurodynamics is constructed. The specific structure of the selected biological neural network is determined according to the specific functions to be achieved and the specific tasks to be completed. Including but not limited to the specific brain areas involved in the task, the number of neurons, neuron types, synaptic types and connection methods. The entire dynamic model system is constructed by digitally simulating the microscopic, mesoscopic and macroscopic neural tissue structures and connection characteristics used by the biological brain to complete the same task.

[0082] The dynamic characteristics of the bioelectric activities generated by neuronal cell bodies and synaptic connections between neurons are expressed in the form of differential equations. A neurodynamic model is constructed through neurodynamic modeling to facilitate the analysis of the neuroimaging data of the experimental subjects.

[0083] Specifically, the above step S202 includes:

[0084] Step S2021, based on the extracted dynamic characteristics, determine the dorsal visual pathway of the biological brain that performs motion perception tasks and several key brain regions corresponding to the dorsal visual pathway, and construct neuron models for each key brain region respectively.

[0085] Step S2022: Based on the differential equations, a neural dynamics model is constructed according to the neuron models of each key brain region. The neural dynamics model is used to simulate the dorsal visual pathway of the biological brain.

[0086] In the embodiment of the present invention, the construction of the neurodynamic model requires extracting the key structures that perform the corresponding functions from the biological brain and performing digital modeling of the key structures based on differential equations.

[0087] Take the construction of a neurodynamic model that realizes the motion perception function and performs the task of judging the direction of moving points as an example. It is necessary to identify a group of signal points with the same motion direction from a group of randomly moving noise points, judge the motion direction of the signal points, and output a decision of left or right, thereby completing the task of judging the direction of motion. This process involves motion perception, evidence accumulation, and competitive decision-making. The loop that performs the "motion perception-evidence accumulation-competitive decision-making" function in the human brain is the dorsal visual pathway, which is composed of four key brain areas, such as Figure 2 As shown, Figure 2 They are all key brain regions in the dorsal visual pathway of the biological brain. Figure 2 and Figure 3 LGN is the lateral geniculate body, V1 is the primary visual cortex, MT is the middle temporal gyrus, and LIP is the lateral wall of the intraparietal sulcus. The model needs to simulate this pathway and construct four corresponding modules. The basic unit of each module is the LIF neuron model, that is, a set of differential equations that simulate the firing of neurons (see the following formula (1)). The overall framework of the model is shown in Figure 3 As shown, Figure 3 The LGN layer is divided into two types of LIF neuron groups, on and off. The MT layer contains two neuron groups, L and R, each with multiple neurons. The V1 layer is composed of multiple neurons and is divided into two groups, G1 and G2. Each V1 neuron receives input current from multiple on-off neurons in a specific arrangement and is connected to the membrane potential output of the LGN layer neurons through the synaptic conductance of the AMPA synapse. The modeling formulas are shown in the following formulas (1)-(5).

[0088]

[0089] I i,AMPA =(U i -U i,AMPA )∑ j g ji,AMPA s j,AMPA

[0090]

[0091] I i,GABA =(U i -U i,GABA )∑ j g ji,GABA s j,GABA

[0092]

[0093] Among them, formula (1) is a single LIF neuron model, where U is the membrane potential of the neuron, τ is the time constant of the neuron, and E L is the membrane potential constant of the neuron in the resting state, I ext The external stimulation current received by the neuron.

[0094] Formula (2) is the receptive field model of LGN neurons, where A is the receptive field coverage, x, y are the horizontal and vertical viewing angles, α, β, All of them are empirical coefficients, and are 1, 1, 0.008, and 0.016 respectively.

[0095] Formula (3) is the input current and conductivity of the AMPA type synapse of LIF neuron, where I i,AMPA is the input current of the AMPA synapse of the i-th neuron, U i is the membrane potential of the ith neuron, U i,AMPA is the membrane potential of the AMPA synapse of the i-th neuron, g ji,AMPA is the AMPA synaptic conductance of the jth neuron to the ith neuron, s j,AMPA is the AMPA synaptic conductance of the jth neuron, τ s,AMPA is the time constant of AMPA synapse, δ is the impulse response function, t k is the time when the kth spike response arrives in the neuron input.

[0096] Formula (4) is the input current and conductivity of the NMDA type synapse of LIF neurons, where I i,NMDA is the input current of the NMDA synapse of the i-th neuron, U i is the membrane potential of the ith neuron, U i,NMDA is the membrane potential of the NMDA synapse of the ith neuron, gji,NMDA is the NMDA synaptic conductance of the jth neuron to the ith neuron, s j,NMDA is the NMDA synaptic conductivity of the jth neuron, τ s,NMDA is the time constant of NMDA synapse, δ is the impulse response function, t k is the time when the kth spike response arrives in the neuron input, [Mg 2+ ] is the empirical value of magnesium ion concentration in synaptic channels, s(t) and x(t) are two types of state variables of NMDA synaptic conductivity, which together constitute the formula for changes in NMDA synaptic conductivity.

[0097] Formula (5) is the input current and conductivity of the GABA type synapse of LIF neurons, where I i,GABA is the input current of the GABA synapse of the i-th neuron, U i is the membrane potential of the ith neuron, U i,GABA is the membrane potential of the GABA synapse of the ith neuron, g ji,GABA is the GABA synaptic conductance of the jth neuron to the ith neuron, s j,GABA is the GABA synaptic conductivity of the jth neuron, τ s,GABA is the time constant of GABA synapse, δ is the impulse response function, t k is the time when the kth spike response arrives in the neuron input.

[0098] Figure 3 The neuron types (on / off / G1 / G2 / R / L / A / B / I) of each module in the model are all applied in the form of the LIF equation group. The LGN layer of the model is divided into two types of LIF neuron groups, on and off, each of which contains multiple groups of LIF neuron models. Both on and off neurons have a double Gaussian spatial receptive field (Formula 2). The viewing angle range covered by the receptive field is defined by x, y (x, y∈[-0.175,0.175]). The pixel area corresponding to the overlap and regular arrangement in space forms a two-dimensional plane corresponding to the image. On and off neurons have different temporal characteristics τ, and on neurons respond slower than off neurons. The input image is spatially convolved with a spatial convolution kernel, and then temporally convolved with a sliding time window as the stimulus-evoked current I of each neuron in the on and off neuron groups. ext , which constitutes the input current of the LGN layer neurons; the membrane potential U changes with time through the differential equation. When the membrane potential reaches a certain threshold, it is considered that the neuron has generated a discharge, and its discharge time t is recorded. k. The models of the other three brain regions are also composed of LIF neuron equations. Each neuron group is connected to the neurons in the previous layer with a specific structure and probability. The V1 layer is composed of multiple neurons and is divided into two groups, G1 and G2. Each V1 neuron receives input current from a specific arrangement of on-off neuron pairs and is connected to the LGN layer neurons through AMPA synapses (Formula 3). Among them, in the LGN projections received by the G1 group neurons, the on cells are always distributed on the left side of the off cells, and in the LGN projections received by the G2 group neurons, the on cells are always distributed on the right side of the off cells ( Figure 5c ). The MT layer contains two neuron groups, L and R, each with multiple neurons. The neurons in the L group receive input currents from the neurons in the G1 group within a certain receptive field, and the neurons in the R group receive input currents from the neurons in the G2 group within a certain receptive field. Both are connected to the membrane potential output of the V1 layer by the same AMPA synaptic conductance. The LIP layer is composed of excitatory neuron groups A and B and inhibitory neuron group I. The neurons in group A receive projections from the neurons in group L, and the neurons in group B receive projections from the neurons in group R. The A and B neuron groups, as well as A, B and inhibitory neuron group I are connected through AMPA synapses and NMDA synapses (Formula 4). The inhibitory neuron group I then inhibits the neurons in groups A and B through GABA synapses (Formula 5) ( Figure 3 ). The connection strength of the above synapses, i.e., the synaptic conductance g, all obey the normal distribution.

[0099] In some optional embodiments, after the neural dynamics model is constructed, the neural dynamics model parameters are initialized. Specifically, the parameters of the neural dynamics model, including but not limited to the specific parameters in the differential equation group, are fitted with reference to the specific data of the biological experiment.

[0100] Specifically, the various parameters of the neurodynamic model, such as the number of connections between neurons, the distribution structure, the mapping relationship between input and output currents, etc., can be preset based on the empirical data of neurophysiological research. Taking the motion perception neurodynamic model as an example, the basic parameters in formulas (1)-(5) refer to the electrophysiological data fitted by the neuronal firing experiment, and the time constant, synaptic conductivity and other parameters are preset a priori.

[0101] Through historical literature retrieval or functional neuroimaging experiments on experimental subjects, the key brain areas corresponding to the dorsal visual pathway of the biological brain that performs motion perception tasks are determined, and brain area division is achieved to facilitate the construction of neuronal models of each key brain area. Based on the set of differential equations, a neural dynamics model is established to complete the simulation of the dorsal visual pathway of the biological brain, so that the neural dynamics model can be analyzed using neuroimaging data.

[0102] Step S203: collect random point motion videos, and input the random point motion videos into the neural dynamics model for performance testing.

[0103] Step S204, comparing the movement direction output by the neural dynamics model with the actual movement direction to measure the sensitivity of the neural dynamics model.

[0104] In the embodiment of the present invention, task data with different types of noise or collected from real environments are used as input data of the neural dynamics model to test the neural dynamics model, and the actual performance of the neural dynamics model in executing the task under the current structure and parameters, such as accuracy, recall, etc., is obtained.

[0105] Specifically, the input, output and performance test methods of the neural dynamics model are based on the modeling objectives of the model and the specific experimental task settings. Taking the motion perception neural dynamics model as an example, its input data is a random point motion video. In a random point motion video, there is a certain proportion of randomly selected points (signal points) moving left or right, and the remaining points (noise points) are randomly distributed. The model receives each frame of video signal as input and gives the output of direction judgment after receiving the video. The model performs performance testing in the random point motion data set, and the output judgment results are compared with the real answers to calculate the judgment accuracy of the model at each difficulty level. The "accuracy-difficulty" curve is fitted using the psychometric curve (Formula 6) to calculate the sensitivity k value for measuring the performance of the model, that is, as a quantitative indicator for judging the performance of the model. The specific calculation method of the k value is: the accuracy p and difficulty data x obtained after running the model for multiple groups of experiments are used to fit the parameters k and b using Formula (6) and least squares estimation. The obtained fitting value k is the sensitivity indicator for evaluating the performance of the model, where the larger the k value, the better the model performance.

[0106]

[0107] Among them, p is the accuracy, x is the difficulty data, and b is the fitting offset.

[0108] By inputting the collected random point motion video into the neural dynamics model, a sensitive performance test of the neural dynamics model is performed to optimize the neural dynamics model, enhance the performance of the neural dynamics model, and enhance the physiological interpretability of the neural dynamics model.

[0109] Step S205, obtaining the scanned neuroimaging data of the experimental subject in the process of performing cognitive behavior.

[0110] For details, please see Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0111] Step S206: extracting structural features and functional features from the neural imaging data.

[0112] Specifically, the above step S206 includes:

[0113] Step S2061, performing brain region segmentation based on the brain structure partition map and the brain function partition map.

[0114] Step S2062, extracting structural features of key brain regions from the neuroimaging data according to the brain structure partitioning map.

[0115] Step S2063, extracting functional features of key brain regions from the neuroimaging data according to the brain function partitioning map.

[0116] In an embodiment of the present invention, taking the experiment of motion perception task as an example, a certain number of healthy organisms are used as experimental subjects, and behavioral experiments and neuroimaging scanning experiments consistent with the neurodynamic model task are completed to obtain brain structure and brain function data information of the experimental subjects.

[0117] Based on the brain structure partition map and brain function partition map, the individual brain is segmented and the key brain areas related to experiments and modeling are delineated. In the example of the motion perception task, the key brain areas are Figure 2 The four key gray matter brain regions and the white matter connecting brain regions between them are shown in Figure 1. Figure 4a , Figure 4b ). In Figure 4, Figure 4a Schematic diagram representing the relationship between the V1-MT white matter connection brain areas and their average FA values ​​and behavioral performance. Figure 4b Schematic diagram of the relationship between MT-LIP inter-white matter connectivity brain regions and their white matter volumes and behavioral performance. Figure 4c Schematic representation of the relationship between the AACIC brain region and its MT-AAIC resting functional connectivity and behavioral performance, Figure 4d Schematic representation of the relationship between the prefrontal suborbital brain area and its LIP-suborbital brain area and behavioral performance.

[0118] The structural features extracted from the imaging data include: gray matter volume (number of voxels) of each key brain region, cortical area, average cortical thickness, etc., white matter volume (number of voxels) of white matter connection areas between key brain regions, average FA value of white matter, average MD value of white matter, etc. Functional features include resting-state functional connectivity between key brain regions, etc. Correlation analysis was performed on these feature data with the behavioral quantitative indicators of the subjects to screen out brain regions and their features that were significantly correlated with the behavioral indicators ( Figure 4b , Figure 4d). In this experimental example, two structural features and two functional features were screened out and were significantly correlated with the behavioral indicators of the subjects ( Figure 4b , Figure 4d ).

[0119] By using the brain structure partition map and the brain function partition map as the basis, brain region segmentation is performed so as to extract structural features and functional features according to the brain structure partition map and the brain function partition map respectively, thereby realizing feature extraction of neuroimaging data.

[0120] Step S207, converting the extracted structural features and functional features into parameters with physiological significance in the pre-built neural dynamics model, and performing parameter tuning on the neural dynamics model.

[0121] Specifically, the above step S207 includes:

[0122] Step S2071 , performing correlation analysis on the extracted structural features and functional features and the pre-acquired behavioral quantitative indicators, and screening out features that are significantly correlated with the behavioral quantitative indicators.

[0123] Step S2072, converting the features that are significantly correlated with the behavioral quantitative indicators into parameters with physiological significance in the neural dynamics model, the parameters including synaptic connection parameters and synaptic connection probability.

[0124] Step S2073, adjusting the synaptic connection parameters and synaptic connection probability in the neural dynamics model.

[0125] In the embodiment of the present invention, taking motion perception as an example, the neuroimaging features related to the behavioral indicators are converted into parameters with corresponding physiological significance in the neurodynamic model, and the parameter values ​​can be explored and adjusted accordingly. During the adjustment process, the model performance is continuously tested. The parameters in the model are adjusted in stages, so that the performance of the model is improved, thereby realizing the tuning of the model performance.

[0126] The parameter adjustment of the model can be reflected in the adjustment of the basic structure of the model, including but not limited to the basic dynamic parameter properties of neurons and synapses, the number of synapses connecting neurons, and the change of connection weights. It can also simulate the response of biological neural networks when directly stimulated by external current by adding other dynamic modules, and adjust the operating state of the model in the form of added current, voltage, etc. to achieve performance improvement.

[0127] Taking the imaging results of the random dot motion experiment as an example, the results of the correlation analysis of human subjects on the key structural features found in the experiment, namely the structural connections of the white matter regions between key brain regions, can qualitatively show that the subjects with denser white matter connections in the region perform better in the behavioral experiment ( Figure 4a , Figure 4b ). Based on this, the synaptic connection parameters g and their connection probabilities (Formulas 4 and 5) between the corresponding modules of the neural dynamics model are adjusted. The synaptic conductance g follows a normal distribution Increase It can increase the conductance of synapses, increase the average connection strength between brain regions, increase the performance of the model, and gradually shorten the decision time. , the model performance gradually deteriorates and the decision time also becomes significantly longer.

[0128] In addition, since the connection from MT to LIP is probabilistic, the neurons of each LIP module are connected to a certain proportion of the neurons of the MT module. When the connection ratio gradually increases from 10%, the performance gradually improves, and it continues to increase beyond a certain proportion and then deteriorates ( Figure 5a , Figure 5b ).

[0129] When the cognitive behavior is motor perception, the extracted structural and functional features are correlated with the pre-acquired behavioral quantitative indicators to screen the features, obtain the features that are significantly correlated with the behavioral quantitative indicators, perform parameter conversion on the features that are significantly correlated with the behavioral quantitative indicators, and adjust the parameters with physiological significance in the neurodynamic model, thereby achieving parameter tuning of the neurodynamic model and making the parameter tuning process have physiological significance, which reflects the efficiency and interpretability of the parameter tuning process.

[0130] Step S2074, when it is detected that the performance of the neural dynamics model has not changed significantly, stop adjusting the parameters of the neural dynamics model.

[0131] In the embodiment of the present invention, the parameter value continues to change, and when the improvement of the performance of the neurodynamic model no longer produces a significant change, the exploration of a larger range of parameter values ​​can be stopped. The parameter adjustment process is consistent with the behavioral performance of human subjects, demonstrating the feasibility and interpretability of neuroimaging experiments to guide model parameter tuning.

[0132] Regarding the key functional features discovered in the experiment, that is, the resting functional connectivity between key brain regions and other regions such as the insula and prefrontal cortex, the correlation analysis results of human subjects can qualitatively reflect that the more relevant a certain brain region of the insula is to MT and the less relevant a certain brain region of the prefrontal cortex is to LIP, the better the subjects perform in behavioral experiments. Based on this, the synaptic conductivity parameters of the corresponding module of the neurodynamic model are adjusted. Since this neurodynamic model does not involve the modeling of other brain regions, previous neurophysiological studies have believed that the influence of these two brain regions on MT and LIP is mainly reflected in the regulation of the strength of their synaptic excitatory connections. Therefore, the resting functional connectivity index is qualitatively correlated with the excitatory synaptic conductance of the MT and LIP modules. Establish a corresponding relationship ( Figure 5c , Figure 5d ). When the synaptic conductance between MT-LIP modules As the excitatory synaptic connection becomes weaker, the model's performance becomes worse and the decision time becomes longer; synaptic conductance As increases, the performance of the model also decreases slightly, but the decision time becomes shorter ( Figure 5c ); When the LIP synaptic conductivity becomes larger, that is, the LIP excitatory synaptic connection becomes weaker, the performance of the model gradually becomes worse ( Figure 5d ). The parameter tuning process is consistent with the behavioral performance of human subjects, demonstrating the feasibility and interpretability of using neuroimaging experiments to guide model parameter tuning.

[0133] Since the number of behavioral features extracted from neuroimaging is relatively low in spatial dimension compared to the number of all parameters involved in the model, the scope and dimension of the parameter optimization space are greatly reduced. In the multidimensional space formed by these parameters, dynamic search methods such as grid search can be used to find the joint optimal solution between multiple feature dimensions and achieve reasonable optimization of model performance.

[0134] The model tuning method based on neuroimaging and cognitive behavior provided in this embodiment ensures that the performance of the neurodynamic model is at a high level by not adjusting the parameters of the neurodynamic model when the performance of the neurodynamic model has not changed significantly.

[0135] In this embodiment, a model tuning device based on neuroimaging and cognitive behavior is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and will not be repeated hereafter. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0136] This embodiment provides a model tuning device based on neuroimaging and cognitive behavior, such as Figure 6 As shown, including:

[0137] The first acquisition module 601 is used to acquire the scanned neuroimaging data of the experimental subject in the process of performing cognitive behavior.

[0138] The feature extraction module 602 is used to extract structural features and functional features from the neuroimaging data.

[0139] The model tuning module 603 is used to convert the extracted structural features and functional features into parameters with physiological significance in the pre-built neural dynamics model, and to tune the parameters of the neural dynamics model so that the pre-built neural dynamics model and the experimental object show the same change trend.

[0140] In some optional embodiments, the device further comprises:

[0141] The second acquisition module is used to obtain historical neuroimaging experimental results and extract dynamic characteristics from the historical neuroimaging experimental results. The dynamic characteristics are used to describe the bioelectric activity characteristics generated by neuronal cell bodies and synaptic connections between neurons.

[0142] The model building module is used to construct a neural dynamics model using differential equations to represent the extracted dynamic characteristics.

[0143] In some optional embodiments, the model building module includes:

[0144] The neuron model building unit is used to determine the dorsal visual pathway of the biological brain that performs motion perception tasks and several key brain areas corresponding to the dorsal visual pathway based on the extracted dynamic characteristics, and to build neuron models of each key brain area respectively.

[0145] The neurodynamic model building unit is used to build a neurodynamic model based on a set of differential equations and according to the neuron models of each key brain area. The neurodynamic model is used to simulate the dorsal visual pathway of the biological brain.

[0146] In some optional embodiments, the device further comprises:

[0147] The acquisition module is used to acquire random point motion videos and input the random point motion videos into the neural dynamics model for performance testing.

[0148] The performance measurement module is used to compare the movement direction output by the neural dynamics model with the actual movement direction to measure the sensitive performance of the neural dynamics model.

[0149] In some optional implementations, the feature extraction module 602 includes:

[0150] The segmentation unit is used to segment brain regions based on the brain structure partition map and the brain function partition map.

[0151] The first extraction unit is used to extract structural features of key brain regions from neuroimaging data according to a brain structure partitioning map, wherein the structural features include gray matter parameters, cortical parameters, and white matter parameters.

[0152] The second extraction unit is used to extract functional characteristics of key brain regions from the neuroimaging data according to the brain function partitioning map, and the functional characteristics include resting-state functional connectivity data between key brain regions.

[0153] In some optional implementations, the model tuning module 603 includes:

[0154] The feature screening unit is used to perform correlation analysis on the extracted structural features and functional features with the pre-acquired behavioral quantitative indicators, and screen out features that are significantly correlated with the behavioral quantitative indicators.

[0155] The parameter conversion unit is used to convert the features that are significantly correlated with the behavioral quantitative indicators into parameters with physiological significance in the neural dynamics model, including synaptic connection parameters and synaptic connection probability.

[0156] The parameter adjustment unit is used to adjust the synaptic connection parameters and synaptic connection probabilities in the neural dynamics model.

[0157] In some optional embodiments, the device further comprises:

[0158] The model stopping tuning module is used to stop adjusting the parameters of the neural dynamics model when it is detected that the performance of the neural dynamics model has not changed significantly.

[0159] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0160] The model tuning device based on neuroimaging and cognitive behavior in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0161] The embodiment of the present invention also provides a computer device having the above Figure 6 The neuroimaging and cognitive-behavioral based model tuning device is shown.

[0162] See also Figure 7 , Figure 7is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 7 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 7 A processor 10 is taken as an example.

[0163] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0164] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0165] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0166] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0167] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Figure 7 The example of connecting through bus is taken in the following.

[0168] The input device 30 can receive input digital or character information and generate key signal input related to user settings and function control of the computer device, such as a touch screen, etc. The output device 40 can include a display device, etc.

[0169] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0170] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A model tuning method based on neuroimaging and cognitive behavior, characterized in that: The method comprises: Obtaining neuroimaging data of scanned experimental subjects while they are performing cognitive behaviors; Extracting structural features and functional features from the neuroimaging data; The extracted structural features and functional features are converted into parameters with physiological significance in a pre-built neural dynamics model, and the parameters of the neural dynamics model are tuned. The pre-built neural dynamics model and the experimental object show the same change trend.

2. The method according to claim 1, characterized in that The method further comprises: Obtaining historical neuroimaging experimental results, and extracting dynamic characteristics from the historical neuroimaging experimental results, wherein the dynamic characteristics are used to describe the bioelectric activity characteristics generated by neuronal cell bodies and synaptic connections between neurons; The extracted dynamic characteristics are represented by differential equations and a neural dynamics model is constructed.

3. The method according to claim 2, characterized in that When the cognitive behavior is motion perception, the extracted dynamic characteristics are expressed by differential equations to construct a neural dynamic model, including: According to the extracted dynamic features, the dorsal visual pathway of the biological brain that performs motion perception tasks and several key brain regions corresponding to the dorsal visual pathway are determined, and the neuron models of each key brain region are constructed respectively; Based on a set of differential equations, a neural dynamics model is constructed according to the neuron models of the key brain regions, and the neural dynamics model is used to simulate the dorsal visual pathway of the biological brain.

4. The method according to claim 2, characterized in that: When the cognitive behavior is motion perception, after constructing the neural dynamics model, the method further includes: Collecting random point motion videos, and inputting the random point motion videos into a neural dynamics model for performance testing; The motion direction output by the neural dynamics model is compared with the actual motion direction to measure the sensitivity of the neural dynamics model.

5. The method according to claim 1, characterized in that When the cognitive behavior is motion perception, the extracting structural features and functional features of the neural imaging data includes: Brain area segmentation is performed based on the brain structure partition map and brain function partition map; Extracting structural features of key brain regions from the neuroimaging data according to the brain structure partitioning atlas, wherein the structural features include gray matter parameters, cortical parameters, and white matter parameters; Functional features of key brain regions are extracted from the neuroimaging data according to the brain functional partitioning map, wherein the functional features include resting-state functional connectivity data between key brain regions.

6. The method according to claim 1, characterized in that When the cognitive behavior is motion perception, the extracted structural features and functional features are converted into parameters with physiological significance in a pre-built neural dynamics model, and the parameters of the neural dynamics model are tuned, including: The extracted structural and functional features are correlated with the previously acquired behavioral quantitative indicators to screen out the features that are significantly correlated with the behavioral quantitative indicators; Converting features that are significantly correlated with behavioral quantitative indicators into parameters with physiological significance in the neural dynamics model, wherein the parameters include synaptic connection parameters and synaptic connection probabilities; Parameter adjustment is performed on synaptic connection parameters and synaptic connection probability in the neural dynamics model.

7. The method according to claim 1, characterized in that The method further comprises: When it is detected that the performance of the neural dynamics model has not changed significantly, the parameter adjustment of the neural dynamics model is stopped.

8. A model tuning device based on neuroimaging and cognitive behavior, characterized in that: The device comprises: An acquisition module, used to acquire the neuroimaging data of the scanned experimental subject during the cognitive behavior; A feature extraction module, used for extracting structural features and functional features from the neural imaging data; The model tuning module is used to convert the extracted structural features and functional features into parameters with physiological significance in a pre-built neural dynamics model, and perform parameter tuning on the neural dynamics model. The pre-built neural dynamics model and the experimental object show the same change trend.

9. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the model tuning method based on neuroimaging and cognitive behavior according to any one of claims 1 to 7 by executing the computer instructions.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the model tuning method based on neuroimaging and cognitive behavior according to any one of claims 1 to 7.

Images