Image Processing Method for Important Brain Regions of Pilots

Abstract

A method for processing pilot's important brain region images, belonging to the technical field of image processing. The purpose of the present invention is a method for processing pilot's important brain region images that extracts pilot's brain images to obtain the characteristics of the flight professional brain, and then uses them to guide the design of the neural network algorithm for autonomous flight unmanned aerial vehicles. The steps of the present invention are: finding the origin, motion correction of the image, spatial normalization, image segmentation, and each line in the file obtained after thresholding is the value linearly related to the gray matter density of each pilot's brain region. The present invention designs an artificial neural network suitable for the flight of unmanned aerial vehicles, so that the operation of unmanned aerial vehicles can be closer to the operation form of flight personnel, increasing the precise flight state of unmanned aerial vehicles, and solving problems such as the robustness, transferability, common sense, and large amount of labeled samples of existing networks.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing. Background Art

[0002] The human brain is a very complex biological system with hundreds of billions of neurons; there are complex nerve fibers connecting at synapses between neurons, forming neural networks and neural circuits that dominate various brain functions. In the brain, every thought will transmit charged signals throughout the brain, and each kind of thought has its neural pathway, which objectively exists, and thoughts will have a certain impact on every cell in the body. Artificial intelligence has always been imitating the information transmission, processing methods, etc. of the human brain to form judgments and decisions.

[0003] The neural network of the brain has a characteristic that is extremely beneficial to adapting to the environment, namely plasticity. Stimuli from the environment, including interactions with the brains of other people, continuously change the neural network of the brain. From the moment of birth, the brain begins to explore the surrounding world and environment, and information also continuously shapes the structure and function of the brain, enabling us to adapt to the needs of the environment. The neural network has high plasticity. Therefore, the electrical activity caused by the environment can dominate the proliferation, consolidation, and pruning of the neural network - preserving appropriate and useful connections and pruning redundant and useless connections. Therefore, the growth experiences of pilots who have undergone many years of flight training are stored in the structure of the neural network. The long-term principle of using it or losing it has shaped different brain structures and network connections, forming a personality and cognitive ability beneficial to flying.

[0004] The human brain has approximately 100 billion neurons, and a single neuron can transmit information through synapses with an average of several other neurons to form neural circuits and pathways. Neurotransmitters are messengers that transmit information between neurons. The synapse at the axon terminal of the previous neuron releases neurotransmitters into the cell gap. When the neurotransmitter receptors at the dendritic synapses of the next neuron receive neurotransmitters of sufficient intensity, they open ion channels, regulate sodium-potassium and other ion channels, convert chemical signals into electrical signals and transmit them inside the neuron, and then the information is transmitted, generating the advanced functions of the brain.

[0005] The transmission and communication of information in the brain are thus realized through electro-chemical signals. These electro-chemical signals travel along brain fibers (nerve axons) and converge at a special node (neuron synapse) of the receiving cell fibers (neuron dendrites) to achieve information transmission. Multiple neurons form neural circuits with different functions.

[0006] Long-term and high-intensity training will lead to changes in the gray matter density, volume, and / or surface variability of the gray matter region in a specific brain region, and may also lead to changes in the amount of white matter in the corpus callosum region. Therefore, the wisdom of a certain professional brain can be transferred to artificial intelligence, enabling artificial intelligence to naturally possess a certain professional quality.

[0007] Current artificial intelligence technologies lack generality. Different models and learning data are used for different tasks, and the same system cannot be used for learning two different tasks. However, the human brain uses the same information processing system for automatic multi-modal perception, information integration, problem analysis and solution, decision-making, and behavior control. The key to the sustainable development of artificial intelligence lies in whether there are breakthroughs in machine learning algorithms, which are upgraded from relying on large amounts of labeled data and high computing power to being able to learn small amounts of unlabeled and multi-modal data, and being as energy-efficient as the human brain. This is the key to transforming from dedicated artificial intelligence to general artificial intelligence. Whoever can first develop these new-generation machine learning algorithms and implement them in computing devices will gain the upper hand in the future field of artificial intelligence.

[0008] Convolutional networks imitate the characteristics of human vision in processing different features in different places, extract different features with different templates, and eliminate the step of artificial feature extraction in machine learning. Human visual perception processing is divided into anterior visual cortex processing, primary visual cortex processing, and high-level visual cortex processing according to anatomical structure: anterior cortex processing includes retinal processing and lateral geniculate nucleus LGN processing; primary visual cortex processing includes SOC filtering processing, layer 6 processing of the visual cortex, layer 4 of the visual cortex, and bipolar cells in layers 2 and 3 of the visual cortex; high-level visual cortex processing includes surface filling in area V4 and surface contour processing in area V3. Summary of the Invention

[0009] The purpose of the present invention is to obtain the characteristics of the flight professional brain by extracting the brain images of pilots, and then use them to guide the image processing method of important brain regions of pilots for the design of the neural network algorithm of autonomous flight drones.

[0010] The present invention uses the main functions of these brain regions, the physical connections between brain regions, and the direction of signal transmission to guide the design of the neural network for the autonomous flight of drones. The steps are as follows:

[0011] S1. Method for finding the origin:

[0012] Open the T1 structural image in Niffit format after format conversion of the pilot, and then refer to the origin position of the MNI152 standard image to accurately find the origin position. At the anterior commissure, remember this position, give an image, align this position, and place the coordinate origin on it;

[0013] S2. Motion correction of the image

[0014] After determining the origin, the line where the anterior commissure and the posterior commissure are located is the Y-axis, and adjust the angle of the nuclear magnetic resonance image of the pilot's head to correct the image direction, that is, head motion correction;

[0015] After rotation through three angles, the relationship between the new coordinates and the coordinates in the original coordinate system:

[0016]

[0017] where: yaw is the yaw angle; pitch is the pitch angle; roll is the roll angle;

[0018] S3. Spatial normalization: Spatial normalization is a registration process. In the registration process, linear and non - linear registrations are carried out. (1) Linear registration is to perform an affine transformation, which is a linear operation. Affine transformation includes rotation, translation, scaling, and shearing. The parameters of rotation, scaling, and translation are determined by comparing the three coordinate planes of the image with the three coordinate planes of the standard image, and then applied to the transformation of the image.

[0019]

[0020] (2) Non - linear registration is to perform distortion and deformation operations on the image after linear registration, aligning the sulci and other structures of the subject image to the same template. After non - linear registration, the brain shapes, tissue shapes, and anatomical positions of different subjects have become consistent, that is, the anatomical positions corresponding to the same spatial coordinates are the same. Non - linear registration mainly projects the three coordinate planes into two - dimensional images, determines the transformation parameters through feature point matching of the two - dimensional images, and then realizes non - linear registration through matrix multiplication.

[0021] S4. Image segmentation

[0022] After spatial normalization, the tissue is segmented into gray matter, white matter, cerebrospinal fluid, and the standardized brain image without the skull. After VBM processing, there are six files in the result folder, namely: the original image, the file recording the correction information, the standardized gray matter map, the standardized white matter map, the file recording the total volumes of gray matter, white matter, and cerebrospinal fluid, and the image after removing the skull, correction, and spatial normalization.

[0023] S5. Value extraction

[0024] Values of each brain region are extracted from the T1 map after removing the brain and standardizing and correcting. After spatial normalization, the brain regions are placed in the respective spaces of the standard brain regions. At this time, the value of each brain region extracted is a value linearly related to the density. The density size indicates the relative volume size of the individual brain region compared to the standard brain region. A larger value indicates that this brain region has become more developed through long - term use. Each line in the file obtained after value extraction is the value linearly related to the gray matter density of each pilot's brain region.

[0025] The present invention collects the brain images of pilots with many years of flight experience in a resting state. After processing and correcting the brain map regions, it obtains the brain regions with significant differences between pilots and ordinary people, as well as the physical connections between brain regions and the sequence of signal transmission, and designs an artificial neural network suitable for UAV flight, so as to make the operation of the UAV closer to the operation form of flight personnel, increase the precise flight state of the UAV, and solve problems such as the robustness, transferability, common sense of the existing network, and the large amount of labeled samples. Brief Description of the Drawings

[0026] Figure 1 is the structural image analysis process;

[0027] Figure 2 is the MNI152 standard brain template;

[0028] Figure 3 is the origin position map;

[0029] Figure 4 is the head motion correction schematic diagram;

[0030] Figure 5 is the linear registration schematic diagram;

[0031] Figure 6 is the comparison diagram of linear registration and non-linear registration images;

[0032] Figure 7 is the standardization schematic diagram;

[0033] Figure 8 is the tissue segmentation schematic diagram;

[0034] Figure 9 is the schematic diagram of the files in the tissue segmentation result folder;

[0035] Figure 10 is the ranking diagram of the brain region changes of pilots compared with ordinary people;

[0036] Figure 11 is the diagram of three eigenvalues and eigenvectors;

[0037] Figure 12 is the scalar index diagram of the diffusion tensor;

[0038] Figure 13 is the diagram of each tensor index image;

[0039] Figure 14 is the coronal, sagittal, and axial diagrams of the fiber tracking results;

[0040] Figure 15 is the connection matrix diagram of the fiber connections between 120 brain regions of 2 randomly selected subjects;

[0041] Figure 16 is the average connectivity matrix graph;

[0042] Figure 17 is a schematic diagram of a neural network;

[0043] Figure 18 This is a network connection diagram of important brain areas of AAL2 pilots;

[0044] Figure 19 It is a computational implementation of a convolutional mask;

[0045] Figure 20 There are three convolution template calculation methods;

[0046] Figure 21 It is a pooling implementation;

[0047] Figure 22 is the activation function graph. DETAILED DESCRIPTION

[0048] The main idea of ​​the present invention is to obtain the brain structure, cognition and functional characteristics of top pilots through the processing and analysis of the most advanced brain science research methods, such as T1 sequence structural images of head nuclear magnetic resonance imaging, diffusion spectrum imaging DSI, ASL, fMRI, DWI, etc., to guide the design of deep neural networks for autonomous flight and combat of drones, that is, intelligent design of flying brains. The present invention uses the main functions of these brain regions, the physical connections between brain regions, and the direction of signal transmission to guide the design of autonomous flight neural networks for drones.

[0049] To establish a computational model that can perform flight cognitive tasks and explain the brain's information processing process, it is necessary to understand the mathematical principles and computational models of brain information processing. There is still a lot of information in the structure and functional mechanism of the brain that has not been explored, which contains huge possibilities. The application of neurosurgery technology in brain science research is becoming more and more mature. Humans have a deeper and continuous understanding of brain science related situations such as the structure and function of the neural circuits of brain cognitive functions, and the information processing mechanism of brain perception and cognitive functions. Brain science and artificial intelligence research have been getting closer and closer. The mutual reference and integration of the two is expected to promote a new round of intelligent technology revolution.

[0050] The experimental design uses the Voxel-based Morphometry (VBM) method to analyze the gray matter characteristics of the T1 sequence data of the pilots and the control group, and then uses the "value enhancement" method to analyze the T1 sequence data of the pilots' brains compared with those of ordinary people to study the special group characteristics of the pilots' brain area models related to vision, movement, somatosensory, visual-spatial processing, and emotional management abilities. This serves as the basis for designing different functional neural network feature models.

[0051] I. Image Feature Extraction

[0052] The T1 - sequence nuclear magnetic resonance (NMR) scan images are a series of images obtained by layer - by - layer scanning in the X, Y, and Z axis directions respectively. Through software, these images can be reconstructed into a brain, that is, the three - dimensional structure of the scan images of a certain sequence of a subject. The conventional experimental method for T1 structural images is to obtain the NMR data of the brain structures of elite pilots and ordinary people, and then perform statistical analysis through a general linear model to find the brain regions with significant differences between them. The experimental process designed is as follows: image inspection, format conversion, determination of the origin, head motion correction, spatial normalization to a common template, tissue segmentation, extraction of voxel values of each brain region, and obtaining the gray matter volume change map of the brain of elite pilots' occupations.

[0053] Software used: matlab2013 and spm8; Experimental data: the structural images of the heads of 296 pilots over 40 years old with many years of flight experience; Parameter selection in the experiment: Tissue Probability Map: TPM.Nii; DartelTemplate: MNI152; The role of the common template is to determine a common coordinate space so that for different brains in this coordinate space, the organizational structures corresponding to the same coordinate positions are basically the same. It is obtained by averaging 152 human brain T1 scan images. Internationally popular head NMR image processing software such as CAT, FSL, and SPM all use the MNI space as the standard template.

[0054] 1. Finding the origin

[0055] Method for finding the origin: Open the T1 structural image in Niffit format after format conversion of the pilot, and then accurately find the origin position with reference to MNI152, at the anterior commissure. Such as Figure 3 the cross - intersection point. Remember this position, given an image, align it with this position and place the coordinate origin on it.

[0056] 2. Motion correction of the image

[0057] After determining the origin, the line where the anterior commissure and the posterior commissure are located is the Y - axis. Adjust the angle of the pilot's head NMR image to correct the image direction, that is, head motion correction. It is to place the head NMR image in a standard coordinate system so as to perform feature analysis with a unified method later.

[0058] yaw

[0059] pitch

[0060] roll

[0061] After rotation by three angles, the relationship between the new coordinates and the original coordinate system coordinates

[0062]

[0063] 3. Spatial normalization

[0064] Spatial normalization is a registration process, and linear and non - linear registrations are carried out during the registration process.

[0065] (1) Linear registration is to perform an affine transformation, which is a linear operation. Affine transformation includes rotation, translation, scaling, and shearing, etc. Affine transformation operates on the whole image, so it can only match the overall position and size. Therefore, there are still large differences in the shape and tissue position of the registered brain.

[0066] Affine transformation includes rotation, translation, and scaling. The parameters of rotation, scaling, and translation are determined by comparing the three coordinate planes of the image with the three coordinate planes of the standard image, and then applied to the transformation of the image.

[0067]

[0068] Non - linear registration is to perform warping and deforming operations on the image after linear registration, aligning the sulci and other structures of the subject image to the same template as much as possible to make it as similar to the template as possible. The operation process is controlled by the optimization theory and some smoothing metrics, and the goal is to minimize the squared error function. After non - linear registration, the brain shapes, tissue shapes, and anatomical positions of different subjects have become very similar - the anatomical positions corresponding to the same spatial coordinates are basically the same.

[0069] Non - linear registration mainly projects the three coordinate planes into two - dimensional images, determines the transformation parameters through the matching of feature points in the two - dimensional images, and then realizes non - linear registration through matrix multiplication.

[0070] Spatial registration seems the same on the surface, but there is also a value recording the relative density of each brain region, which can be converted into values of density and volume.

[0071] 4. Image segmentation

[0072] After spatial normalization, the tissue is segmented into gray matter, white matter, cerebrospinal fluid, and the standardized brain without the skull. After VBM processing, there are six files in the result folder, namely: the original image, the file recording the correction information, the standardized gray matter map, the standardized white matter map, the file recording the total volumes of gray matter, white matter, and cerebrospinal fluid (unit: cubic centimeter), and the map after removing the skull & correction & spatial normalization.

[0073] The brain - removed & standardized T1 map with the subject name prefixed by wmr. In the experiment of this project, the data prefixed by wmr was used to reduce errors.

[0074] 5. Value extraction

[0075] Values of each brain region are proposed for the T1 map after brain peeling and standardized correction. Since after spatial standardization, the brain regions are placed in the respective spaces of the standard brain regions, the value of each brain region proposed at this time is a value linearly related to the density. The size of the density indicates the relative volume size of the individual brain region compared to the standard brain region. A larger value indicates that this brain region has become more developed through long-term use.

[0076] Each line in the file obtained after value extraction is the value linearly related to the gray matter density of 116 brain regions of each pilot. Table 1 is a partial set of values intercepted after the file is opened. Through these values, the volume change of the corresponding brain region can be calculated. Therefore, it can be said that by applying the size of these values, the magnitude of the impact of flight training on a certain brain region can be obtained. By averaging and sorting the relative density values of each brain region of pilots in different regions, the ranking of the impact of flight training on each brain region of pilots in that region can be obtained. By averaging the relative density values of each brain region of all pilots, the relative density values of 116 brain regions of all pilots are obtained. The size of the value represents the magnitude of the change in the volume of this brain region of the pilot compared to ordinary people as the flight training progresses, such as Figure 10 。

[0077] Table 1 Results after value extraction

[0078]

[0079] Figure 10 On the horizontal axis is the number of 116 gray matter brain regions, and on the vertical axis is the magnitude of the change in the brain of elite pilots shaped by long-term flight training compared to the ordinary person's brain. The part greater than zero is the part where the brain region volume increases. The larger the value, the greater the degree of volume increase. Through this ranking, the flight-related qualities of the pilots can be obtained.

[0080] According to the order of the ranking and integrating brain regions with similar functions, the flight-related qualities of the pilots are obtained: According to the order of the degree of increase in brain region volume, and then through the main functions of the brain regions, the main characteristics of the pilots' cognition and behavior are obtained. The main flight qualities ranked from strong to weak are: ① Physical regulation and motor function; ② Psychological construction and emotion processing; ③ Multi-sensory linkage and situation awareness; ④ Visual discrimination and spatial orientation; ⑤ Fine motor skills and information processing; ⑥ Learning comprehension and episodic memory; ⑦ Target recognition and judgment-making; ⑧ Advanced cognition and self-evaluation.

[0081] II. Connectivity analysis of diffusion magnetic resonance imaging

[0082] 1. Imaging principle

[0083] Assuming that the diffusion of water molecules conforms to a three-dimensional Gaussian distribution, through derivation, the following formula is obtained:

[0084] D is a 3×3 symmetric matrix, and the positive definite matrix D is the tensor matrix.

[0085]

[0086] Where: The eigenvalues λ1≥λ2≥λ3, as Figure 11 shown, the eigenvectors v i ⊥v j , i≠j.

[0087] 2. Related scalar indices of the diffusion tensor

[0088] The relationship between the related scalar indices of the diffusion tensor is shown as Figure 12 shown.

[0089] ① Mean Diffusivity (MD)

[0090]

[0091] ② Fractional Anisotropy (FA)

[0092]

[0093] ③ Axial Diffusion (AD)

[0094] AD = λ1

[0095] ④ Radial Diffusivity (RD)

[0096]

[0097] The images of each index are as Figure 13 shown.

[0098] 3. Experimental analysis process

[0099] ① Data preparation

[0100] Select the DSI images from the Dicom files of the head MRI images of 74 pilots over 40 years old with many years of flight experience, and then perform format conversion - dicom to NIFTI.

[0101] After conversion, each folder contains three files simultaneously: ".bval" file, ".bvec" file, and ".nii.gz" file. The ".bval" file stores the b-values for each scanning direction. Generally, the first b = 0, and common values for the subsequent b are 1000, 2400, etc.; the ".bvec" file stores the x, y, z direction vectors for each scanning direction, and the sum of the squares of the three directions should be 1.

[0102] ② The fiber tracking results, such as Figure 14 shown.

[0103] ③ The connection matrix of the fiber connection situation between brain regions, such as Figure 15 shown.

[0104] The average connection matrix obtained by averaging the connection matrices of 74 people, such as Figure 16 shown.

[0105] III. Implications for the Design of Deep Neural Networks

[0106] Artificial intelligence networks are designed by imitating the human brain. The design includes network structure design, algorithm design, and network training process design. An intelligent network is composed of several input layer nodes, output layer nodes, and several hidden layer nodes (as shown in Figure 17 ).

[0107] Use a connection diagram to represent the connection matrix between the brain regions with prominent changes in pilots to obtain the logical relationship between brain regions.

[0108] The ultimate development direction of artificial neural networks is to simulate the entire brain. For the current CNN (Convolutional Neural Network), it is inspired by the biological visual system and simulates the information processing patterns of multiple levels of the brain. CNN is the most widely used convolutional neural network in the current field of computer vision. According to different purposes, a part of Figure 18 can be taken as an artificial neural network. If the previous figures are organized and sorted according to the CNN network, the left and right thalamus can be used as input nodes, and the subsequent nodes are stratified according to the weight size.

[0109] A complete CNN generally consists of a convolutional layer, a pooling layer, a transmission layer, a fully connected layer, etc. The basis is the orientation column of the human visual eye: a group of neurons that respond preferentially to lines and edges of similar angles form an orientation column. All the cortical cells in each column act as a functional module to process the input from a certain position in the visual field and transmit the processed information to other regions. Different convolution templates obtain different features after operation for subsequent calculations. For nodes with known functions, some filters are imitated from the human brain to reduce the amount of computation and improve the accuracy.

[0110] IV. Algorithm Design

[0111] ① Design idea of feature extraction algorithm (convolution template).

[0112] Different convolution templates obtain different features after operation, which are used for subsequent calculations. For nodes with known functions, some filters can be imitated from the human brain to reduce the amount of calculation and improve the accuracy rate.

[0113] ② Design simplified extraction, that is, an algorithm to combine multiple neurons into one node (pooling is shown in the following figure. As Figure 21 shown, the method of extracting the maximum number in the area is adopted, that is, the maximum value of the feature is retained.)

[0114] Design neuron activation function.

[0115] To avoid gradient explosion and computational complexity, the RELU function is used as the activation function for transmission.

[0116]

[0117] V. Visualization Design Environment

[0118] The visualization of neural networks mainly includes the visualization of input and output, weights, activation functions, pooling, etc. When designing, some visualization methods can be used to intuitively see the running process of the data of the designed network, which is convenient for quickly adjusting the network design and can help speed up the design process. The common methods are as follows:

[0119] ① Tensorflow-Playground is a graphical online demonstration and experimental platform for simple neural networks for teaching purposes, with very powerful visualization of neural network structures and network training processes.

[0120] ② To make it more convenient for Tensor Flow programmers to understand, debug and optimize, Google has released a set of integrated visualization tools called Tensor Board. Tensor Board can be used to display Tensor Flow images, draw quantitative index graphs generated by images and additional data.

[0121] ③ Netscope is an online visualization tool that supports the neural network structure of Caffe.

[0122] ④ Deep convolutional networks are commonly used neural networks in the field of image processing. Deep convolutional networks have achieved great performance breakthroughs in many pattern recognition tasks. High-quality deep models rely on a large number of attempts. Applying the visualization analysis system CNNVis can help better understand, analyze and design deep convolutional networks.

[0123] ⑤deep-visualization-toolbox

[0124] ⑥It is also possible to write your own program to create an interface that displays the variables of the outputs of the neurons in the neural network being trained, the variables of the weights, the intermediate results, etc. every second or at certain time intervals.

Claims

1. An image processing method for important brain regions of pilots, characterized in that: Using the main functions of these brain regions, the physical connections between brain regions, and the direction of signal transmission to guide the design of the neural network for the autonomous flight of drones, the image processing steps for brain regions are as follows: S1. Method for finding the origin: Open the Niffit-format T1 structural image of the pilot after format conversion, and then refer to the origin position of the MNI152 standard image to accurately find the origin position. At the anterior commissure, remember this position, give an image, align it with this position, and place the coordinate origin on it; S2. Motion correction of the image After determining the origin, the line where the anterior commissure and the posterior commissure are located is the Y-axis. Adjust the angle of the nuclear magnetic resonance image of the pilot's head to correct the image direction, that is, head motion correction; After three-angle rotation, the relationship between the new coordinates and the original coordinate system coordinates: Where: yaw is the yaw angle; pitch is the pitch angle; roll is the roll angle; S3. Spatial normalization: Spatial normalization is a registration process. During the registration process, linear and non-linear registrations are required. (1) Linear registration is to perform an affine transformation, which is a linear operation. The affine transformation includes rotation, translation, scaling, and shearing. The parameters of rotation, scaling, and translation are determined by comparing the three coordinate planes of the image with the three coordinate planes of the standard image, and then applied to the transformation of the image (2) Non-linear registration is to perform distortion and deformation operations on the image after linear registration, aligning the brain sulci and other structures of the subject image with the same template. After non-linear registration, the brain shapes, tissue shapes, and anatomical positions of different subjects have become consistent, that is, the anatomical positions corresponding to the same spatial coordinates are the same; non-linear registration mainly projects the three coordinate planes into two-dimensional images, determines the transformation parameters through feature point matching of the two-dimensional images, and then realizes non-linear registration through matrix multiplication; S4. Image segmentation After spatial normalization, the tissue is segmented into gray matter, white matter, cerebrospinal fluid, and the standardized brain image without the skull. After VBM processing is completed, there are six files in the result folder, namely: the original image, the file recording the correction information, the standardized gray matter map, the standardized white matter map, the file recording the total volumes of gray matter, white matter, and cerebrospinal fluid, and the image after removing the skull, correction, and spatial normalization; S5. Value extraction Extract the values of each brain region from the T1 image after removing the brain and standardizing and correcting. After spatial normalization, the brain regions are placed in the respective spaces of the standard brain regions. At this time, the value of each brain region extracted is a value linearly related to the density. The density size indicates the relative volume size of the individual brain region compared to the standard brain region. A larger value indicates that this brain region has become more developed through long-term use; each line in the file obtained after value extraction is the value linearly related to the gray matter density of each pilot's brain region.

Images