Brain feature recognition method and device, computer equipment and storage medium

Abstract

The invention relates to a brain feature recognition method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a first brain region feature matrix before image vision training and a second brain region feature matrix after image vision training of a target user; determining a first candidate brain region feature sequence of the target user before image vision training and a second candidate brain region feature sequence of the target user after image vision training according to a feature difference quantized value between the first brain region feature matrix and the second brain region feature matrix; and according to the first candidate brain region feature sequence and the second candidate brain region feature sequence, determining brain change features of the target user in the image vision training process. By adopting the method, the brain change characteristics of the imaging physician in the training process can be accurately identified.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to a brain feature recognition method, device, computer equipment and storage medium. Background Art

[0002] With the development of medical technology, medical imaging technology plays a key role in the field of modern medicine. When processing medical images, image acquisition relies on advanced technology, while image interpretation is the responsibility of imaging physicians.

[0003] In the process of training radiologists, it is usually necessary to provide novice radiologists with special imaging vision training. In related technologies, targeted imaging vision training can be developed for radiologists based on the characteristic changes in the central mechanism of the brain during the imaging vision training of radiologists, so as to improve their ability to interpret medical images.

[0004] Based on this, how to accurately identify the characteristics of brain changes of radiologists during training has become a technical problem that needs to be solved urgently. Summary of the invention

[0005] Based on this, it is necessary to provide a brain feature recognition method, device, computer equipment and storage medium that can accurately identify the brain change characteristics of imaging physicians during training to address the above technical problems.

[0006] In a first aspect, the present application provides a method for identifying brain features, the method comprising:

[0007] Obtaining a first brain region feature matrix of the target user before image vision training and a second brain region feature matrix after image vision training;

[0008] Determine, according to a quantized value of a feature difference between a first brain region feature matrix and a second brain region feature matrix, a first candidate brain region feature sequence of the target user before image vision training and a second candidate brain region feature sequence after image vision training;

[0009] According to the first candidate brain region feature sequence and the second candidate brain region feature sequence, brain change characteristics of the target user during the image vision training process are determined.

[0010] In the technical solution provided by the embodiment of the present application, on the basis of obtaining the first brain region feature matrix of the target user before image vision training and the second brain region feature matrix after image vision training, according to the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix, the first candidate brain region feature sequence of the target user before image vision training and the second candidate brain region feature sequence after image vision training are determined, and then according to the first candidate brain region feature sequence and the second candidate brain region feature sequence, the brain change characteristics of the target user during image vision training are determined. In this method, the first brain region feature matrix of the target user before image vision training and the second brain region feature matrix after image vision training are compared, and the feature difference quantization value is used to evaluate the feature change amplitude of each brain region of the target user before and after training, so as to distinguish the significant change features and the insignificant change features in the brain region feature matrix before and after training, and provide an objective basis for the subsequent determination of the first candidate brain region feature sequence and the second candidate brain region feature sequence, improve the accuracy and effectiveness of the first candidate brain region feature sequence and the second candidate brain region feature sequence, and then ensure the effectiveness of the brain change characteristics identified based on the candidate brain region feature sequence. In addition, during the brain feature recognition process, a phased approach is adopted to gradually screen the candidate brain region feature sequences and brain change features of the target user. While ensuring the effectiveness of the features, it reduces the redundancy of the features and improves the efficiency of brain feature recognition to a certain extent.

[0011] In one embodiment, determining the brain change characteristics of the target user during the image vision training process according to the first candidate brain region feature sequence and the second candidate brain region feature sequence includes:

[0012] According to the first candidate brain region feature sequence and the second candidate brain region feature sequence, a preset brain region state classification model is trained to obtain an initial classification accuracy value;

[0013] Based on the initial classification accuracy value, updating the features of the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence;

[0014] The high-resolution features of the first candidate brain region feature sequence and the high-resolution features of the second candidate brain region feature sequence are determined as brain change features.

[0015] In the technical solution provided in the embodiment of the present application, a preset brain region state classification model is trained according to the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an initial classification accuracy value, and based on the initial classification accuracy value, the first candidate brain region feature sequence and the second candidate brain region feature sequence are updated to gradually eliminate low-resolution features in the candidate brain region feature sequences that have a low contribution to model classification, so as to obtain more accurate brain change characteristics of the target user during image vision training.

[0016] In one embodiment, based on the initial classification accuracy value, the first candidate brain region feature sequence and the second candidate brain region feature sequence are updated to obtain high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence, including:

[0017] Perform feature deletion on the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an updated feature sequence set;

[0018] Inputting the updated feature sequence set into the brain region state classification model to obtain an updated classification accuracy value of the updated feature sequence set;

[0019] When the updated classification accuracy value is greater than the initial classification accuracy value, the features in the first candidate brain region feature sequence and the second candidate brain region feature sequence and not in the updated feature sequence set are determined as low-resolution features;

[0020] The low-resolution features are removed from the first candidate brain region feature sequence to obtain the high-resolution features of the first candidate brain region feature sequence, and the low-resolution features are removed from the second candidate brain region feature sequence to obtain the high-resolution features of the second candidate brain region feature sequence.

[0021] In the technical solution provided in the embodiment of the present application, the first candidate brain region feature sequence and the second candidate brain region feature sequence are subjected to feature deletion to obtain an updated feature sequence set, and the updated feature sequence set is input into the brain region state classification model to obtain an updated classification accuracy value of the updated feature sequence set. When the updated classification accuracy value is greater than the initial classification accuracy value, the features that are in the first candidate brain region feature sequence and the second candidate brain region feature sequence and are not in the updated feature sequence set are determined as low-resolution features, and are removed from each candidate brain region feature sequence to obtain high-resolution features. This feature selection method of gradually eliminating the features that contribute the least to the model performance to achieve the optimal feature subset not only significantly improves the accuracy of feature screening and the efficiency of model training, but also enhances the generalization ability and robustness of the model, providing a more accurate and efficient solution for complex data sets such as magnetic resonance imaging.

[0022] In one embodiment, the target user includes multiple users; obtaining a first brain region feature matrix of the target user before image vision training and a second brain region feature matrix after image vision training includes:

[0023] Collecting a first brain image of each target user before image vision training and a second brain image after image vision training;

[0024] For each target user, extract a first feature matrix of a first brain image of the target user in multiple brain regions, and extract a second feature matrix of a second brain image of the target user in multiple brain regions;

[0025] The first feature matrices of the target users are fused to obtain a first brain region feature matrix; and the second feature matrices of the target users are fused to obtain a second brain region feature matrix.

[0026] In the technical solution provided in the embodiment of the present application, first feature matrices of first brain images of multiple target users before image vision training in multiple brain regions are collected, and the first feature matrices of each target user are fused to obtain a first brain region feature matrix, and, first feature matrices of first brain images of multiple target users before image vision training are collected in multiple brain regions, and the first feature matrices of each target user are fused to obtain a first brain region feature matrix, which is equivalent to uniformly processing the features of multiple target users to support unified recognition of brain features of multiple target users and improve the application scenarios of brain feature recognition methods.

[0027] In one embodiment, collecting a first brain image of each target user before image vision training and a second brain image of each target user after image vision training includes:

[0028] Acquire a first brain image sequence and a second brain image sequence of a plurality of candidate users;

[0029] Performing head motion correction on the first brain image sequence and the second brain image sequence of each candidate user to determine the head motion range of each candidate user;

[0030] Determine a candidate user whose head movement range is less than a preset movement threshold as a target user;

[0031] A first brain image of each target user is determined from a first brain image sequence of each target user, and a second brain image of each target user is determined from a second brain image sequence of each target user.

[0032] In the technical solution provided in the embodiment of the present application, for any candidate user, a first brain image sequence and a second brain image sequence of the candidate user are obtained, and then the head movement range of each candidate user is determined. When the head movement range is less than a preset movement threshold, the candidate user is determined as a target user, and the first brain image and the second brain image of the target user are determined from the first brain image sequence and the second brain image sequence. In this way, the method for determining brain images proposes users with relatively active brain states, so that the brain images of the target users are all collected when the brain state is relaxed, thereby improving the accuracy and effectiveness of the brain images.

[0033] In one embodiment, extracting a first feature matrix of a first brain image of a target user in multiple brain regions, and extracting a second feature matrix of a second brain image of a target user in multiple brain regions, includes:

[0034] Obtaining a low-frequency amplitude feature matrix, a local consistency feature matrix, and a functional connection feature matrix of the first brain image in multiple brain regions, and fusing the low-frequency amplitude feature matrix, the local consistency feature matrix, and the functional connection feature matrix to obtain a first feature matrix;

[0035] The low-frequency amplitude feature matrix, local consistency feature matrix and functional connection feature matrix of the second brain image in multiple brain regions are obtained, and the low-frequency amplitude feature matrix, local consistency feature matrix and functional connection feature matrix are fused to obtain a second feature matrix.

[0036] In the technical solution provided in the embodiment of the present application, the features of the first brain image and the second brain image are extracted from three feature dimensions: low-frequency amplitude, local consistency, and functional connectivity features. The first feature matrix and the second feature matrix determined in this way have richer expressive capabilities and can more accurately reflect the feature changes of the target user before and after image vision training.

[0037] In one embodiment, determining a first candidate brain region feature sequence of a target user before image vision training and a second candidate brain region feature sequence after image vision training according to a feature difference quantization value between a first brain region feature matrix and a second brain region feature matrix includes:

[0038] Calculate the quantitative values ​​of feature differences corresponding to the first brain region feature matrix and the second brain region feature matrix in different brain regions;

[0039] For a feature difference quantization value of any brain region, if the feature difference quantization value is greater than a preset difference quantization threshold, determining a brain region identifier corresponding to the feature difference quantization value;

[0040] The brain region features matching the brain region identifier in the first brain region feature matrix are determined as the first candidate brain region feature sequence of the target user, and the brain region features matching the brain region identifier in the second brain region feature matrix are determined as the second candidate brain region feature sequence of the target user.

[0041] In the technical solution provided in the embodiment of the present application, the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix is ​​calculated, and the significant change and non-significant change features in the brain region feature matrix are distinguished from the feature level, which provides an objective quantitative basis for subsequent feature screening. Then, when the feature difference quantization value is greater than the preset difference quantization threshold, the brain region identifier corresponding to the feature difference quantization value is determined, and then the first candidate brain region feature sequence matching the brain region identifier in the first brain region feature matrix and the second candidate brain region feature sequence matching the brain region identifier in the second brain region feature matrix are determined, so as to improve the accuracy and effectiveness of the first candidate brain region feature sequence and the second candidate brain region feature sequence.

[0042] In a second aspect, the present application also provides a brain feature recognition device, comprising:

[0043] A data acquisition module, used to acquire a first brain region feature matrix of a target user before image vision training and a second brain region feature matrix after image vision training;

[0044] A feature screening module, used to determine a first candidate brain region feature sequence of the target user before the image vision training and a second candidate brain region feature sequence after the image vision training according to a feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix;

[0045] The feature recognition module is used to determine the brain change characteristics of the target user during the image vision training process based on the first candidate brain region feature sequence and the second candidate brain region feature sequence.

[0046] In a third aspect, the present application further provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the method in any one of the embodiments of the first aspect are implemented.

[0047] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method in any one of the embodiments of the first aspect above.

[0048] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method in any one of the embodiments of the first aspect above. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the drawings required for use in the embodiments of the present application or related technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 is an internal structure diagram of a computer device in one embodiment;

[0051] Figure 2 is a schematic diagram of a flow chart of a brain feature recognition method in one embodiment;

[0052] Figure 3 It is a flowchart of the steps of obtaining a brain feature matrix in one embodiment;

[0053] Figure 4 is a flowchart of a brain image acquisition step in one embodiment;

[0054] Figure 5 It is a flowchart of the steps of obtaining a brain feature sequence in one embodiment;

[0055] Figure 6 A schematic diagram of a flow chart of a step of acquiring brain change characteristics in one embodiment;

[0056] Figure 7 A schematic diagram of a process of obtaining high-resolution features in one embodiment;

[0057] Figure 8 A schematic flow chart of a high-resolution feature acquisition step in another embodiment;

[0058] Fig. 9 is a flow chart of a crowd classification method in one embodiment;

[0059] Fig.10 4 is a structural block diagram of a brain feature recognition device in one embodiment. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0061] With the development of medical technology, medical imaging technology plays a key role in the field of modern medicine, from the initial X-rays and ultrasound to the current digitalization, to reveal more detailed information about the structure and function of the body.

[0062] In actual application scenarios, when processing medical images, although image acquisition relies on advanced technology, image interpretation still needs to be the responsibility of radiologists. Therefore, it is very important to provide specialized image vision training for novice radiologists to improve their ability to interpret medical images.

[0063] Based on this, the embodiments of the present application provide a brain feature recognition method, device, computer equipment and storage medium, which objectively identify the biomarker characteristics related to the professional visual ability of the imaging physician by identifying the characteristic changes of the brain central mechanism during the imaging physician's imaging vision training, so as to reveal the neurological basis for the formation of the imaging physician's professional ability, and then formulate targeted imaging vision training for the imaging physician. At the same time, it also provides a scientific basis for the development of more efficient professional talent training programs, which has important theoretical and practical significance.

[0064] The brain feature recognition method provided in the embodiment of the present application can be applied to a computer device. The computer device can be a server, and its internal structure diagram can be as follows: Figure 1 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store brain feature recognition data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a brain feature recognition method is implemented.

[0065] Those skilled in the art will understand that Figure 1 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0066] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0067] In an exemplary embodiment, Figure 2 As shown, a method for identifying brain features is provided, and the method is described by taking the method applied to a computer device as an example. The computer device can be a terminal or a server, and can also be applied to a computer system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In the embodiment of the present application, the method for identifying brain features includes the following steps:

[0068] S201, obtaining a first brain region feature matrix of a target user before image vision training and a second brain region feature matrix after image vision training.

[0069] Among them, image vision training refers to professional training courses designated to improve users' ability to recognize and interpret images, and target users refer to users who participate in image vision training, which can be one user or a user group including multiple users.

[0070] In actual applications, the process of image vision training for target users is also a process in which the characteristics of the target users' brain central mechanisms change. By influencing visual training for target users, the target users' cognitive abilities are further improved.

[0071] In order to reveal the impact of image vision training on the brain center of the target user, the embodiment of the present application adopts a magnetic resonance scanning method to collect brain images of the target user before image vision training, and performs feature extraction of the region of interest on the brain features of the target user to obtain a first brain region feature matrix of the target user; after the image vision training is completed, the magnetic resonance scanning method is again adopted to collect brain images of the target user, and feature extraction of the region of interest is performed on the brain features of the target user to obtain a second brain region feature matrix of the target user.

[0072] S202, determining a first candidate brain region feature sequence of the target user before the image vision training and a second candidate brain region feature sequence after the image vision training according to the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix.

[0073] The first brain region feature matrix is ​​used as the brain features of each brain region before the target user undergoes image vision training, and the second brain region feature matrix is ​​used as the brain features of each brain region after the target user undergoes image vision training. The first brain region feature matrix and the second brain region feature matrix have the same size and represent the same brain region.

[0074] The feature difference quantification value between the first brain region feature matrix and the second brain region feature matrix is ​​calculated to determine the degree of differentiation between the features of the first brain region feature matrix and the second brain region feature matrix in the same brain region.

[0075] It can be understood that the number of feature difference quantization values ​​is consistent with the number of features in the first brain region feature matrix and the number of brain regions in the first brain region feature matrix, and a feature difference quantization value represents the degree of differentiation between the feature corresponding to a brain region in the first brain region feature matrix and the feature corresponding to the brain region in the second brain region feature matrix.

[0076] For the feature difference quantification value between the two features corresponding to each brain region, the larger the feature difference quantification value, the more significant the feature change of the brain region, and the more it can represent the changes in the brain centers of the target user during the image vision training process. Conversely, the smaller the feature difference quantification value, the less significant the feature change of the brain region, and the less able to represent the changes in the brain centers of the target user during the image vision training process.

[0077] Based on this, a feature difference quantization threshold can be set in advance, and the brain region corresponding to the feature difference quantization value greater than the feature difference quantization threshold is determined as the target brain region, and the features corresponding to the first brain region feature matrix in the target brain region are determined as the first candidate brain region feature sequence of the target user, and the features corresponding to the second brain region feature matrix in the target brain region are determined as the second candidate brain region feature sequence of the target user.

[0078] S203, determining brain change characteristics of the target user during the image vision training process according to the first candidate brain region feature sequence and the second candidate brain region feature sequence.

[0079] After determining the first candidate brain region feature sequence and the second candidate brain region feature sequence, the features in the first candidate brain region feature sequence and the second candidate brain region feature sequence can be directly determined as brain change features to characterize the features that change significantly during the training process of the target user. Alternatively, effective feature pairs in the first candidate brain region feature sequence and the second candidate brain region feature sequence can be further screened according to the feature selection strategy to obtain more accurate brain change features.

[0080] Optionally, the first candidate brain region feature sequence and the second candidate brain region feature sequence are subjected to mean-variance normalization processing, and then the normalized first candidate brain region feature sequence and the second candidate brain region feature sequence are differenced element by element, and features whose difference is less than a preset threshold are removed from the first candidate brain region feature sequence and the second candidate brain region feature sequence, so as to obtain high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence, i.e., brain change features.

[0081] Optionally, a sequential backward selection method is used to eliminate features in the first candidate brain region feature sequence and the second candidate brain region feature sequence one by one or in batches until a preset number of features is reached, thereby obtaining high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence, i.e., brain change features.

[0082] In the embodiment of the present application, on the basis of obtaining the first brain region feature matrix of the target user before the image vision training and the second brain region feature matrix after the image vision training, the first candidate brain region feature sequence of the target user before the image vision training and the second candidate brain region feature sequence after the image vision training are determined according to the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix, and then the brain change characteristics of the target user during the image vision training are determined according to the first candidate brain region feature sequence and the second candidate brain region feature sequence. In this method, the first brain region feature matrix of the target user before the image vision training and the second brain region feature matrix after the image vision training are compared, and the feature difference quantization value is used to evaluate the feature change amplitude of each brain region of the target user before and after the training, so as to distinguish the significant change features and the insignificant change features in the brain region feature matrix before and after the training, and provide an objective basis for the subsequent determination of the first candidate brain region feature sequence and the second candidate brain region feature sequence, improve the accuracy and effectiveness of the first candidate brain region feature sequence and the second candidate brain region feature sequence, and then ensure the effectiveness of the brain change characteristics identified based on the candidate brain region feature sequence. In addition, during the brain feature recognition process, a phased approach is adopted to gradually screen the candidate brain region feature sequences and brain change features of the target user. While ensuring the effectiveness of the features, it reduces the redundancy of the features and improves the efficiency of brain feature recognition to a certain extent.

[0083] Next, a feasible method for obtaining the brain region feature matrix in the aforementioned embodiment is described.

[0084] In an exemplary embodiment, Figure 3 As shown, the target users include multiple ones; obtaining the first brain region feature matrix of the target users before the image vision training and the second brain region feature matrix after the image vision training includes:

[0085] S301 , collecting a first brain image of each target user before image vision training and a second brain image of each target user after image vision training.

[0086] The first brain image and the second brain image may be resting-state functional magnetic resonance imaging (rs-fMRI).

[0087] Before each target user undergoes image vision training, a first brain image of each target user is collected by means of magnetic resonance scanning. After each target user undergoes image vision training, a second brain image of each target user is collected again by means of magnetic resonance scanning.

[0088] It should be noted that the first brain image and the second brain image include a functional image and an anatomical image. After the first brain image and the second brain image are acquired, the images may be preprocessed.

[0089] Optionally, the functional image and anatomical image of each target user are registered to remove the linear trend in the time series; then the registered image of the target user is normalized to the MNI space and resampled to 3 mm isotropic voxels, and finally a 0.01–0.08 Hz temporal filtering is performed on the time series of each voxel to reduce the influence of low-frequency drift and high-frequency noise, thereby obtaining the preprocessed first brain image of each target user and the preprocessed second brain image of each target user.

[0090] S302: For each target user, extract a first feature matrix of a first brain image of the target user in multiple brain regions, and extract a second feature matrix of a second brain image of the target user in multiple brain regions.

[0091] Based on the preset brain atlas, the first brain image and the second brain image of each target user are respectively segmented into regions of interest to obtain different brain regions of the first brain image and different brain regions of the second brain image, wherein the brain regions include but are not limited to different regions of the cerebral cortex, key nodes of the functional network and other regions.

[0092] Based on multiple feature scales, feature extraction is performed on each brain region (i.e., the segmented region of interest) in the first brain image to obtain feature matrices of different feature scales, and the feature matrices at multiple feature scales are fused to obtain a first feature matrix of the first brain image; feature extraction is performed on each brain region (i.e., the segmented region of interest) in the second brain image to obtain feature matrices of different feature scales, and the feature matrices at multiple feature scales are fused to obtain a second feature matrix of the second brain image.

[0093] In an exemplary embodiment, extracting a first feature matrix of a first brain image of a target user in multiple brain regions, and extracting a second feature matrix of a second brain image of a target user in multiple brain regions, includes:

[0094] Obtaining a low-frequency amplitude feature matrix, a local consistency feature matrix, and a functional connection feature matrix of the first brain image in multiple brain regions, and fusing the low-frequency amplitude feature matrix, the local consistency feature matrix, and the functional connection feature matrix to obtain a first feature matrix;

[0095] The low-frequency amplitude feature matrix, local consistency feature matrix and functional connection feature matrix of the second brain image in multiple brain regions are obtained, and the low-frequency amplitude feature matrix, local consistency feature matrix and functional connection feature matrix are fused to obtain a second feature matrix.

[0096] Different feature dimension extraction models are called to perform feature extraction on the first brain image and the second brain image to obtain a first feature matrix corresponding to the first brain image and a second feature matrix corresponding to the second brain image.

[0097] Optionally, the first brain image is input into a low-frequency amplitude recognition model to obtain a low-frequency amplitude feature matrix of the first brain image in multiple brain regions; the first brain image is input into a local consistency recognition model to obtain a local consistency feature matrix of the first brain image in multiple brain regions; the first brain image is input into a functional connectivity recognition model to obtain a functional connectivity feature matrix of the first brain image in multiple brain regions, and then the low-frequency amplitude feature matrix, local consistency feature matrix and functional connectivity feature matrix output by the three models are weightedly summed element by element to obtain the first feature matrix.

[0098] In another implementation scenario, the low-frequency amplitude feature matrix, the local consistency feature matrix, and the functional connectivity feature matrix output by the three models may be stacked to obtain a first feature matrix.

[0099] Taking the first brain image including N regions of interest as an example, the sizes of the low-frequency amplitude feature matrix, the local consistency feature matrix and the functional connectivity feature matrix are all 1×N. The above three matrices are stacked to obtain a 3×N matrix, namely the first feature matrix.

[0100] Optionally, a low-frequency amplitude feature matrix of the second brain image in multiple brain regions is extracted through a low-frequency amplitude recognition model; a local consistency feature matrix of the second brain image in multiple brain regions is extracted through a local consistency recognition model; a functional connectivity feature matrix of the second brain image in multiple brain regions is extracted through a functional connectivity recognition model, and then the low-frequency amplitude feature matrix, the local consistency feature matrix and the functional connectivity feature matrix are weightedly summed element by element to obtain a second feature matrix.

[0101] In another implementation scenario, the low-frequency amplitude feature matrix, local consistency feature matrix, and functional connectivity feature matrix output by the three models may be stacked to obtain a second feature matrix.

[0102] Taking the second brain image including N regions of interest as an example, the sizes of the low-frequency amplitude feature matrix, the local consistency feature matrix and the functional connectivity feature matrix are all 1×N. The above three matrices are stacked to obtain a 3×N matrix, namely the second feature matrix.

[0103] In the embodiment of the present application, the features of the first brain image and the second brain image are extracted from three feature dimensions: low-frequency amplitude, local consistency, and functional connectivity features. The first feature matrix and the second feature matrix determined in this way have richer expressive capabilities and can more accurately reflect the feature changes of the target user before and after image vision training.

[0104] S303, fusing the first feature matrices of the target users to obtain a first brain region feature matrix; and fusing the second feature matrices of the target users to obtain a second brain region feature matrix.

[0105] The first feature matrix of each target user is converted into a one-dimensional row vector, and the row vectors of each target user are stacked to obtain the first brain region feature matrix; the second feature matrix of each target user is converted into a one-dimensional row vector, and the row vectors of each target user are stacked to obtain the second brain region feature matrix.

[0106] Continuing with the example of a brain image including N regions of interest, the size of the first feature matrix of each target user is 3×N, and the size of the second feature matrix of each target user is 3×N, if the number of target users is 30, the first feature matrix of each target user is reduced in dimension to obtain a 1×3N row vector, and then the row vectors of the 30 target users are concatenated to obtain a 30×3N first brain region feature matrix; similarly, the second feature matrix of each target user is reduced in dimension to obtain a 1×3N row vector, and then the row vectors of the 30 target users are concatenated to obtain a 30×3N second brain region feature matrix.

[0107] In an embodiment of the present application, first feature matrices of first brain images of multiple target users before image vision training in multiple brain regions are collected, and the first feature matrices of each target user are fused to obtain a first brain region feature matrix, and first feature matrices of first brain images of multiple target users before image vision training are collected in multiple brain regions, and the first feature matrices of each target user are fused to obtain a first brain region feature matrix, which is equivalent to uniformly processing the features of multiple target users to support unified recognition of brain features of multiple target users and improve the application scenarios of brain feature recognition methods.

[0108] As can be seen from the above embodiments, the brain feature recognition method supports processing brain images of multiple target users. Then, in order to ensure the validity and consistency of the brain images collected by the target users, the properties of the brain images of the target users need to be basically consistent. For example, the brain images of different target users need to be collected in a state of brain relaxation. In this case, it is necessary to eliminate the brain images of users with more active brain states. In an exemplary embodiment, Figure 4 As shown, collecting a first brain image of each target user before image vision training and a second brain image after image vision training includes:

[0109] S401 , obtaining a first brain image sequence and a second brain image sequence of a plurality of candidate users.

[0110] The first brain image sequence is an image sequence collected by the candidate user before the image vision training, including multiple MRI images. The second brain image sequence is an image sequence collected by the candidate user after the image vision training, including multiple MRI images.

[0111] In a feasible scenario, in order to eliminate the magnetization inhomogeneity effect and the environmental adaptability of the candidate user, the brain image sequence of each candidate user can be preprocessed. For example, in the time series, the first 10 MRI images of the first brain image sequence of each candidate user are deleted, and the inter-slice acquisition delay is corrected to obtain an updated first brain image sequence. In addition, the first 10 MRI images of the second brain image sequence of each candidate user are deleted, and the inter-slice acquisition delay is corrected to obtain an updated second brain image sequence.

[0112] S402 , performing head motion correction on the first brain image sequence and the second brain image sequence of each candidate user to determine the head motion range of each candidate user.

[0113] For each set of brain image sequences of each candidate user, the following steps are performed in sequence: determine the head activity center of the brain image with the earliest acquisition time and the head activity center of the brain image with the latest acquisition time in each set of brain image sequences, then calculate the movement distance between the two head activity centers to obtain each head movement range of the candidate user in each set of brain image sequences, and finally select the largest head movement range from multiple head movement ranges of the candidate user as the head movement range of the candidate user.

[0114] S403: Determine the candidate user whose head movement range is smaller than a preset movement threshold as the target user.

[0115] The preset motion threshold may be determined based on an empirical value, such as 0.2 mm, and is used to measure the user's brain activity state during brain image acquisition.

[0116] If the head movement range is greater than the preset movement threshold, it means that the target user's brain state is relatively active during the process of collecting brain images, and the authenticity of the corresponding collected brain images is lower; if the head movement range is less than the preset movement threshold, it means that the target user's brain state is relatively calm during the process of collecting brain images, and the authenticity of the corresponding collected brain images is higher.

[0117] Based on this, candidate users whose head movement range exceeds a preset movement threshold are determined as invalid users, and candidate users whose head movement range is less than the preset movement threshold are determined as target users.

[0118] S404: Determine a first brain image of each target user from the first brain image sequence of each target user, and determine a second brain image of each target user from the second brain image sequence of each target user.

[0119] A brain image is randomly determined from the first brain image sequence of each target user as the first brain image of each target user, and a brain image is randomly determined from the second brain image sequence of each target user as the second brain image of each target user. It should be emphasized that the number of the first brain image in the first brain image sequence is consistent with the number of the second brain image in the second brain image sequence.

[0120] Exemplarily, the initial brain data set includes 100 brain images of user A before image vision training and 100 brain images of user A after image vision training. Then, when user A is determined to be the target user, the 50th of the 100 brain images of user A before image vision training can be used as the first brain image, and the 50th of the 100 brain images of user A after image vision training can be used as the second brain image.

[0121] In an embodiment of the present application, for any candidate user, a first brain image sequence and a second brain image sequence of the candidate user are obtained, and then the head movement range of each candidate user is determined. When the head movement range is less than a preset movement threshold, the candidate user is determined as a target user, and the first brain image and the second brain image of the target user are determined from the first brain image sequence and the second brain image sequence. This method of determining brain images proposes users with relatively active brain states, so that the brain images of the target users are all collected when the brain state is relaxed, thereby improving the accuracy and effectiveness of the brain images.

[0122] After determining the first brain region feature matrix and the second brain region feature matrix of the target user, the feature levels of the two can be compared to screen out features that have significant changes between the first brain region feature matrix and the second brain region feature matrix. Figure 5 As shown, according to the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix, determining the first candidate brain region feature sequence of the target user before the image vision training and the second candidate brain region feature sequence after the image vision training, including:

[0123] S501, calculating a quantitative value of feature difference between a first brain region feature matrix and a second brain region feature matrix.

[0124] It is emphasized again that the first brain region feature matrix and the second brain region feature matrix are feature matrices of the target user in different states, that is, the first brain region feature matrix and the second brain region feature matrix have the same size and the brain region identifiers corresponding to the same matrix position are the same.

[0125] Based on this, a paired t-test was used to compare the features of the first brain region and the second brain region to obtain a feature difference matrix with the same size as the feature matrix of the first brain region. Each matrix element in the feature difference matrix (that is, each feature difference quantification) is used to quantify the change in the features before and after training at a certain brain region of the target user.

[0126] S502: For a feature difference quantization value of any brain region, if the feature difference quantization value is greater than a preset difference quantization threshold, determine a brain region identifier corresponding to the feature difference quantization value.

[0127] The preset difference quantization threshold may be set based on an empirical value, such as 0.05, and is used to measure changes in brain region characteristics of multiple brain regions of the target user before and after image vision training.

[0128] If the feature difference quantization value is greater than the preset difference quantization threshold, it means that the greater the change in the characteristics of the brain area position corresponding to the feature difference quantization value during the training process, the greater the impact of the image vision training on the characteristics of the brain area; if the feature difference quantization value is less than the preset difference quantization threshold, it means that the smaller the change in the characteristics of the brain area position corresponding to the feature difference quantization value during the training process, the smaller the impact of the image vision training on the characteristics of the brain area.

[0129] Based on this, the brain area with a feature difference quantization value greater than the preset difference quantization threshold is determined as the brain area of ​​interest, and the brain area identification is recorded. In other words, determining the brain area identification is equivalent to determining the range of the brain area where the target user has undergone significant feature changes during the image vision training process.

[0130] S503, determining the brain region features matching the brain region identifier in the first brain region feature matrix as a first candidate brain region feature sequence of the target user, and determining the brain region features matching the brain region identifier in the second brain region feature matrix as a second candidate brain region feature sequence of the target user.

[0131] When one or more brain region identifiers are determined, brain region features that match the brain region identifiers are determined from the first brain region feature matrix as the first candidate brain region feature sequence for the target user before image vision training; brain region features that match the brain region identifiers are determined from the second brain region feature matrix as the second candidate brain region feature sequence for the target user after image vision training.

[0132] In the embodiment of the present application, the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix is ​​calculated to distinguish the significant change and non-significant change features in the brain region feature matrix from the feature level, providing an objective quantitative basis for subsequent feature screening. Then, when the feature difference quantization value is greater than the preset difference quantization threshold, the brain region identifier corresponding to the feature difference quantization value is determined, and then the first candidate brain region feature sequence matching the brain region identifier in the first brain region feature matrix and the second candidate brain region feature sequence matching the brain region identifier in the second brain region feature matrix are determined to improve the accuracy and effectiveness of the first candidate brain region feature sequence and the second candidate brain region feature sequence.

[0133] After performing feature differentiation evaluation on the first brain region feature matrix and the second brain region feature matrix, the brain regions corresponding to the first candidate brain region features and the second candidate brain region features can be determined as key brain regions, and the feature pairs of key brain regions can be determined as brain change features during image vision training. The first and second candidate brain region features can also be further screened to enhance the effectiveness of the brain change features.

[0134] In an exemplary embodiment, Figure 6As shown, according to the first candidate brain region feature sequence and the second candidate brain region feature sequence, determining the brain change features of the target user during the image vision training process includes:

[0135] S601, training a preset brain region state classification model according to the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an initial classification accuracy value.

[0136] According to the first candidate brain region feature sequence and the second candidate brain region feature sequence, a preset brain region state classification model is trained to obtain an initial classification accuracy value.

[0137] The brain region state classification model is used to predict the brain region state of a user according to the brain region characteristics of the user, and the brain region state includes before and after the image vision training. Optionally, the brain region state classification model is a support vector machine model (Support Vector Machine, SVM).

[0138] The first candidate brain region feature sequence of the target user and the state label of the first candidate brain region feature sequence (before training) are used as a set of training data, and the second candidate brain region feature sequence and the state label of the second candidate brain region feature sequence (after training) are used as a set of training data to construct a data set. The data set is used as a training set and input into the brain region state classification model for training to obtain the initial classification accuracy value of the brain region state classification model on the data set.

[0139] Optionally, in order to balance the dimensional differences between features and ensure the fairness of model training, the first candidate brain region feature sequence and the second candidate brain region feature sequence can be respectively subjected to mean variance normalization processing before being input into the brain region state classification model, and then the processed first candidate brain region feature sequence and the state label of the first candidate brain region feature sequence (before training) are used as a set of training data, and the processed second candidate brain region feature sequence and the state label of the second candidate brain region feature sequence (after training) are used as a set of training data to generate a data set.

[0140] The execution steps of mean-variance normalization are as follows: first calculate the mean and standard deviation of each feature, and then use these statistics to convert each feature value into a standardized form, that is, each feature value minus its mean and divided by its standard deviation, eliminating the dimensional influence between different features and making the weight of each feature in the model more fair.

[0141] S602, based on the initial classification accuracy value, updating the features of the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence.

[0142] In a feasible application scenario, a preset classification accuracy threshold can be set in advance. If the initial classification accuracy value is greater than the classification accuracy threshold, it means that the features in the first candidate brain region feature sequence and the second candidate brain region feature sequence are valid. At this time, the high-resolution features of the first candidate brain region feature sequence and the high-resolution features of the second candidate brain region feature sequence can be further screened out according to the preset feature update strategy.

[0143] Optionally, based on the feature difference quantization value between the first candidate brain region feature sequence and the second candidate brain region feature sequence, in order of the feature difference quantization value from small to large, features corresponding to the feature difference quantization value are eliminated one by one from the first candidate brain region feature sequence and the second candidate brain region feature sequence, and a new first candidate brain region feature sequence and a new second candidate brain region sequence are generated to obtain a new training set, and then the preset brain region state classification model is trained according to the new training set to obtain a new classification accuracy value; if the new classification accuracy value is still greater than the classification accuracy threshold, continue to perform the above-mentioned step of eliminating features one by one until the new classification accuracy value is equal to or less than the classification accuracy threshold for the first time, stop feature updating, and determine the features in the latest generated first candidate brain region feature sequence as high-resolution features, and determine the features in the latest generated second candidate brain region feature sequence as high-resolution features.

[0144] S603: Determine the high-resolution features of the first candidate brain region feature sequence and the high-resolution features of the second candidate brain region feature sequence as brain change features.

[0145] In an embodiment of the present application, a preset brain region state classification model is trained according to the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an initial classification accuracy value, and based on the initial classification accuracy value, the first candidate brain region feature sequence and the second candidate brain region feature sequence are updated to gradually eliminate low-resolution features in the candidate brain region feature sequence that have a low contribution to model classification, so as to obtain more accurate brain change characteristics of the target user during the image vision training process.

[0146] In an exemplary embodiment, another implementation method of the aforementioned embodiment step S602 "updating the first candidate brain region feature sequence and the second candidate brain region feature sequence based on the initial classification accuracy value to obtain the high-resolution features of the first candidate brain region feature sequence and the high-resolution features of the second candidate brain region feature sequence" is also provided, such as Figure 7 As shown, including:

[0147] S701, performing feature deletion on the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an updated feature sequence set.

[0148] The features of the first candidate brain region feature sequence and the second candidate brain region feature sequence in the same brain region are deleted to obtain an updated feature sequence set.

[0149] S702, inputting the updated feature sequence set into the brain region state classification model to obtain an updated classification accuracy value of the updated feature sequence set.

[0150] The update feature sequence set and the brain region state labels of each update feature sequence in the update sequence set are summarized as a training set and input into the brain region state classification model to obtain the update classification accuracy value obtained by training the brain region state classification model for the update feature sequence set.

[0151] S703: When the updated classification accuracy value is greater than the initial classification accuracy value, the features that are in the first candidate brain region feature sequence and the second candidate brain region feature sequence and are not in the updated feature sequence set are determined as low-resolution features.

[0152] If the updated classification accuracy value is greater than the initial classification accuracy value, it means that the feature deleted in step S701, that is, the feature in the first candidate brain region feature sequence and the second candidate brain region feature sequence and not in the updated feature sequence set, contributes less to the performance of the brain region state classification model, and the feature is determined as a low-resolution feature.

[0153] It should be noted that steps S701 to S703 are steps for deleting a brain region feature and performing resolution determination on it. In practical applications, it is necessary to perform resolution determination on all features in the first candidate brain region feature sequence, which means that it is necessary to repeat the above steps S701 to S703 for multiple times to determine whether each feature in the candidate brain region feature sequence is a high-resolution feature or a low-resolution feature.

[0154] S704, removing low-resolution features from the first candidate brain region feature sequence to obtain high-resolution features of the first candidate brain region feature sequence, and removing low-resolution features from the second candidate brain region feature sequence to obtain high-resolution features of the second candidate brain region feature sequence.

[0155] Each time a low-resolution feature is determined, the low-resolution feature is removed from the first candidate brain region feature sequence, which is equivalent to retaining the high-resolution features in the first candidate brain region feature sequence. Simultaneously, the low-resolution feature is removed from the second candidate brain region feature sequence, and the high-resolution features in the second candidate brain region feature sequence are retained.

[0156] See also Figure 8 , Figure 8Taking the brain region state classification model as a support vector machine as an example, a flow diagram of high-resolution features is provided. Taking the first candidate brain region feature sequence and the second candidate brain region feature sequence including m features as an example, the m features in the first candidate brain region feature sequence and the second candidate brain region feature sequence are used as feature sets and input into the support vector machine to obtain an initial classification accuracy value, then feature 1 is removed from all feature sets to generate a new feature set to obtain a first classification accuracy value, feature 2 is removed from all feature sets to generate a new feature set to obtain a second classification accuracy value, ..., feature m is removed from all features to generate a new feature set to obtain an mth classification accuracy value. Then, the initial classification accuracy value is compared with the first classification accuracy value, the second classification accuracy value, ..., and the mth classification accuracy value, respectively, to determine a target classification accuracy greater than the initial classification accuracy value, and then determine a target feature set corresponding to the target classification accuracy, and the features eliminated in the generation process of the target training set are determined as low-resolution features, and then the low-resolution features are deleted from the feature set to obtain high-resolution features.

[0157] In the embodiment of the present application, the first candidate brain region feature sequence and the second candidate brain region feature sequence are subjected to feature deletion to obtain an updated feature sequence set, and the updated feature sequence set is input into the brain region state classification model to obtain an updated classification accuracy value of the updated feature sequence set. When the updated classification accuracy value is greater than the initial classification accuracy value, the features that are in the first candidate brain region feature sequence and the second candidate brain region feature sequence and are not in the updated feature sequence set are determined as low-resolution features, and are removed from each candidate brain region feature sequence to obtain high-resolution features. This feature selection method of gradually eliminating the features that contribute the least to the model performance to achieve the optimal feature subset not only significantly improves the accuracy of feature screening and the efficiency of model training, but also enhances the generalization ability and robustness of the model, providing a more accurate and efficient solution for complex data sets such as magnetic resonance imaging.

[0158] In an exemplary embodiment, a method for identifying brain features is provided, comprising the following steps:

[0159] (1) obtaining a first brain image sequence and a second brain image sequence of multiple candidate users;

[0160] (2) performing head motion correction on the first brain image sequence and the second brain image sequence of each candidate user to determine the head motion range of each candidate user;

[0161] (3) Determine the candidate user whose head movement range is less than the preset movement threshold as the target user;

[0162] (4) Determining a first brain image of each target user from the first brain image sequence of each target user, and determining a second brain image of each target user from the second brain image sequence of each target user.

[0163] (5) for each target user, extracting a first feature matrix of a first brain image of the target user in multiple brain regions, and extracting a second feature matrix of a second brain image of the target user in multiple brain regions;

[0164] (6) The first feature matrices of the target users are fused to obtain a first brain region feature matrix; and the second feature matrices of the target users are fused to obtain a second brain region feature matrix.

[0165] (7) Calculating the quantitative value of the feature difference between the feature matrix of the first brain region and the feature matrix of the second brain region;

[0166] (8) For the feature difference quantization value of any brain region, if the feature difference quantization value is greater than a preset difference quantization threshold, determine the brain region identifier corresponding to the feature difference quantization value;

[0167] (9) The brain region features that match the brain region identifiers in the first brain region feature matrix are determined as the first candidate brain region feature sequence of each target user, and the brain region features that match the brain region identifiers in the second brain region feature matrix are determined as the second candidate brain region feature sequence of each target user.

[0168] The above steps (7) to (9) are the first stage feature selection process.

[0169] (10) Perform mean-variance normalization on the first candidate brain region feature sequence and the second candidate brain region feature sequence.

[0170] The above step (10) is the feature selection process of the second stage.

[0171] (11) training a preset brain region state classification model according to the processed first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an initial classification accuracy value;

[0172] (12) performing feature deletion on the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an updated feature sequence set;

[0173] (13) inputting the updated feature sequence set into the brain region state classification model (support vector machine) to obtain an updated classification accuracy value of the updated feature sequence set;

[0174] (14) When the updated classification accuracy value is greater than the initial classification accuracy value, the features in the first candidate brain region feature sequence and the second candidate brain region feature sequence and not in the updated feature sequence set are determined as low-resolution features;

[0175] The above steps (12) to (14) are executed cyclically to identify each feature in the first candidate brain region feature sequence as a high-resolution feature or a low-resolution feature.

[0176] (15) The low-resolution features are removed from the first candidate brain region feature sequence to obtain the high-resolution features of the first candidate brain region feature sequence, and the low-resolution features are removed from the second candidate brain region feature sequence to obtain the high-resolution features of the second candidate brain region feature sequence.

[0177] (16) The high-resolution features of the first candidate brain region feature sequence and the high-resolution features of the second candidate brain region feature sequence are determined as brain change features.

[0178] The above steps (11) to (16) are the feature selection process of the third stage.

[0179] In the embodiment of the present application, a phased and systematic feature selection strategy is adopted. First, the paired t-test is used to identify the features with significant statistical differences between the two groups of data, and then the mean variance is normalized to balance the dimensional differences between the features and ensure the fairness of model training. Finally, the sequential backward selection combined with the support vector machine (SBS-SVM) technology is used to gradually eliminate the features that contribute the least to the model performance to achieve the optimal feature subset. Compared with a single feature selection technology, this three-stage feature selection method not only significantly improves the accuracy of feature screening and the efficiency of model training, but also enhances the generalization ability and robustness of the model, providing a more accurate and efficient solution for complex data such as magnetic resonance imaging.

[0180] In order to further verify the effectiveness of the brain change features determined by the brain feature recognition method provided in the embodiment of the present application, the extracted brain change features are input into the SVM classification model to verify the classification performance of the SVM classification model on the feature set.

[0181] After feature screening is completed, the linear support vector machine classifier with soft interval separation strategy and hinge loss function is used to classify the screened features. Among them, soft interval separation allows some samples to be on or outside the interval boundary, which helps to deal with noise and overlap in actual data. The hinge loss function is used to measure the prediction error of the classifier. By minimizing the hinge loss, the optimal classification hyperplane is determined, thereby achieving effective classification.

[0182] In order to evaluate the performance of the constructed classifier, the Leave-One-Out Cross-Validation (LOOCV) method was used. In LOOCV, one sample is used as the test set each time, and the remaining samples are used as the training set. This process is repeated until each sample is used as the test set once. This method can make full use of limited sample data and evaluate the generalization ability of the classifier on unknown data. By calculating indicators such as classification accuracy, recall rate, F1 score, etc., we can fully understand the performance of the classifier. The following is the specific calculation formula of the indicators:

[0183] (1) Classification accuracy: Classification accuracy refers to the ratio of the number of samples correctly classified by the classifier to the total number of samples. The calculation formula is as follows:

[0184]

[0185] (2) Recall: Recall refers to the ratio between the number of positive samples correctly identified by the classifier and the total number of actual positive samples. The calculation formula is as follows:

[0186]

[0187] (3) Precision: Precision refers to the proportion of samples predicted by the classifier to be positive that are actually positive. The calculation formula is as follows:

[0188]

[0189] Among them, TP (True Positive) represents the number of positive samples correctly classified as positive, TN (True Negative) represents the number of negative samples correctly classified as negative, FP (False Positive) represents the number of negative samples incorrectly classified as positive, and FN (False Negative) represents the number of positive samples incorrectly classified as negative.

[0190] The test conclusions obtained through the above verification method can show that the brain change features extracted by this application perform well on the SVM classification model and can be applied to the classification scenario of potential imaging physicians. In addition, the brain state features extracted by this application also improve the stability and generalization performance of the model to a certain extent, thereby providing scientific and effective technical support for the professional training of imaging physicians.

[0191] In an exemplary embodiment, the brain change features provided by the present application can also be applied to other classification scenarios, such as the classification scenario of surgeons, used to evaluate surgeons as interns or surgical experts, or the classification scenario of car evaluators, used to determine whether the evaluators are junior evaluators or senior evaluation experts, etc. In each classification scenario, the flow chart of the classification method is as follows: Fig. 9 As shown, the following steps are included:

[0192] S901, obtain two groups of resting-state functional magnetic resonance imaging datasets from the same population but in different cognitive states.

[0193] Before completing professional visual training, magnetic resonance data sets of each user were collected. After completing professional visual training, magnetic resonance data of each user were collected again, and two sets of resting state functional magnetic resonance imaging data sets from the same population but in different cognitive states were obtained.

[0194] S902, data preprocessing.

[0195] After image acquisition, the image format was first converted to suit subsequent analysis. Subsequently, the first 10 images of each time series were deleted to eliminate the effect of magnetization inhomogeneity and consider the environmental adaptation of the participants. Next, the acquired data were corrected for inter-slice acquisition delays and head motion correction was performed, and participants with head motion exceeding a threshold of 0.2 mm were removed. For the remaining subjects, a two-sample t-test was used to verify whether there was a significant difference in head motion. Subsequently, the average functional data were registered with the anatomical images of the subjects, and the linear trend in the time series was removed. All functional images were then normalized to MNI space and resampled to 3 mm isotropic voxels. Finally, a 0.01–0.08 Hz temporal filtering was performed on the time series of each voxel to reduce the influence of low-frequency drift and high-frequency noise.

[0196] S903, extracting feature matrices at different feature scales.

[0197] The selected brain atlas is used to segment the brain regions of interest in the preprocessed fMRI images. These regions of interest include different areas of the cerebral cortex, key nodes of the functional network, etc. For each segmented region of interest, low-frequency amplitude, local consistency, and resting-state functional connectivity features are extracted. For each feature scale, a 1×N-dimensional feature matrix is ​​obtained, where N is the number of regions of interest.

[0198] S904, fusing the extracted feature matrices at different feature scales to obtain feature fusion matrices of two groups of people in different statuses.

[0199] S905, using a three-stage feature selection method to determine the high-resolution features with the highest discrimination.

[0200] S906, using a support vector machine model to classify the selected high-resolution features.

[0201] The brain feature recognition method provided in the embodiment of the present application is intended to extract and maximize the use of functional imaging features that can reflect the brain state. Through precise brain region segmentation and feature extraction, and the implementation of a strict feature selection process, features that are significantly significant for the brain state classification of the target person are screened out, so as to formulate a personalized training program for the target person, carry out targeted reinforcement for the deficiencies in the training, and thus comprehensively improve the professional quality of the target person.

[0202] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0203] Based on the same inventive concept, the embodiment of the present application also provides a brain feature recognition device for implementing the above-mentioned brain feature recognition method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more brain feature recognition device embodiments provided below can refer to the limitations on the brain feature recognition method above, and will not be repeated here.

[0204] In an exemplary embodiment, Fig.10 As shown, a brain feature recognition device is provided, comprising: a data acquisition module 1001, a feature screening module 1002 and a feature recognition module 1003, wherein:

[0205] The data acquisition module 1001 is used to acquire the first brain region feature matrix of the target user before the image vision training and the second brain region feature matrix after the image vision training;

[0206] The feature screening module 1002 is used to determine a first candidate brain region feature sequence of the target user before the image vision training and a second candidate brain region feature sequence after the image vision training according to the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix;

[0207] The feature recognition module 1003 is used to determine the brain change characteristics of the target user during the image vision training process according to the first candidate brain region feature sequence and the second candidate brain region feature sequence.

[0208] In an exemplary embodiment, the feature recognition module 1003 includes: a model recognition unit, a resolution determination unit and a feature determination unit, wherein:

[0209] A model recognition unit is used to train a preset brain region state classification model according to the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an initial classification accuracy value;

[0210] A resolution determination unit, configured to update the first candidate brain region feature sequence and the second candidate brain region feature sequence based on the initial classification accuracy value, to obtain high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence;

[0211] The feature determination unit is used to determine the high-resolution features of the first candidate brain region feature sequence and the high-resolution features of the second candidate brain region feature sequence as brain change features.

[0212] In an exemplary embodiment, the resolution determination unit includes: a feature deletion subunit, a model evaluation subunit, a low-resolution feature determination subunit and a low-resolution feature elimination subunit, wherein:

[0213] A feature deletion subunit is used to perform feature deletion on the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an updated feature sequence set;

[0214] A model evaluation subunit, used for inputting the updated feature sequence set into the brain region state classification model to obtain an updated classification accuracy value of the updated feature sequence set;

[0215] A low-resolution feature determination subunit, configured to determine, when the updated classification accuracy value is greater than the initial classification accuracy value, a feature that is in the first candidate brain region feature sequence and the second candidate brain region feature sequence and is not in the updated feature sequence set as a low-resolution feature;

[0216] The low-resolution feature elimination subunit is used to eliminate the low-resolution features from the first candidate brain region feature sequence to obtain the high-resolution features of the first candidate brain region feature sequence, and to eliminate the low-resolution features from the second candidate brain region feature sequence to obtain the high-resolution features of the second candidate brain region feature sequence.

[0217] In an exemplary embodiment, the target user includes multiple; the data acquisition module 1001 includes an image acquisition unit, a matrix extraction unit and a matrix fusion unit, wherein:

[0218] An image acquisition unit, used to acquire a first brain image of each target user before image vision training and a second brain image after image vision training;

[0219] A matrix extraction unit, for extracting, for each target user, a first feature matrix of a first brain image of the target user in multiple brain regions, and a second feature matrix of a second brain image of the target user in multiple brain regions;

[0220] The matrix fusion unit is used to fuse the first feature matrices of each target user to obtain a first brain region feature matrix; and to fuse the second feature matrices of each target user to obtain a second brain region feature matrix.

[0221] In an exemplary embodiment, the image acquisition unit includes: a data set acquisition subunit, a head correction subunit, a user screening subunit and an image determination subunit, wherein:

[0222] A data set acquisition subunit, used to acquire a first brain image sequence and a second brain image sequence of a plurality of candidate users;

[0223] A head correction subunit, used to perform head motion correction on the first brain image sequence and the second brain image sequence of each candidate user, and determine the head motion range of each candidate user;

[0224] A user screening subunit, used to determine a candidate user whose head movement range is less than a preset movement threshold as a target user;

[0225] The image determination subunit is used to determine the first brain image of each target user from the first brain image sequence of each target user, and to determine the second brain image of each target user from the second brain image sequence of each target user.

[0226] In an exemplary embodiment, the matrix extraction unit is used to obtain the low-frequency amplitude feature matrix, local consistency feature matrix and functional connectivity feature matrix of the first brain image in multiple brain regions, fuse the low-frequency amplitude feature matrix, local consistency feature matrix and functional connectivity feature matrix to obtain a first feature matrix; and obtain the low-frequency amplitude feature matrix, local consistency feature matrix and functional connectivity feature matrix of the second brain image in multiple brain regions, fuse the low-frequency amplitude feature matrix, local consistency feature matrix and functional connectivity feature matrix to obtain a second feature matrix.

[0227] In an exemplary embodiment, the feature screening module 1002 includes: a quantitative value calculation unit, a brain region identification acquisition unit and a feature matching unit, wherein:

[0228] A quantization value calculation unit, used for calculating a quantization value of a feature difference between a first brain region feature matrix and a second brain region feature matrix;

[0229] A brain region identification acquisition unit is used to determine the brain region identification corresponding to the feature difference quantization value for any brain region if the feature difference quantization value is greater than a preset difference quantization threshold;

[0230] The feature matching unit is used to determine the brain region features that match the brain region identifier in the first brain region feature matrix as the first candidate brain region feature sequence of the target user before the image vision training; and to determine the brain region features that match the brain region identifier in the second brain region feature matrix as the second candidate brain region feature sequence of the target user after the image vision training.

[0231] Each module in the above-mentioned brain feature recognition device can be implemented in whole or in part by software, hardware and their combination. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the corresponding operations of each of the above modules.

[0232] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0233] In an exemplary embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0234] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0235] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0236] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, an artificial intelligence (AI) processor, etc., but are not limited to this.

[0237] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0238] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A method for identifying brain features, characterized in that: The method comprises: Obtaining a first brain region feature matrix of the target user before image vision training and a second brain region feature matrix after image vision training; Determining a first candidate brain region feature sequence of the target user before image vision training and a second candidate brain region feature sequence after image vision training according to a feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix; According to the first candidate brain region feature sequence and the second candidate brain region feature sequence, the brain change characteristics of the target user during the image vision training process are determined.

2. The method according to claim 1, characterized in that The determining, according to the first candidate brain region feature sequence and the second candidate brain region feature sequence, the brain change features of the target user during the image vision training process includes: According to the first candidate brain region feature sequence and the second candidate brain region feature sequence, a preset brain region state classification model is trained to obtain an initial classification accuracy value; Based on the initial classification accuracy value, updating the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence; The high-resolution features of the first candidate brain region feature sequence and the high-resolution features of the second candidate brain region feature sequence are determined as the brain change features.

3. The method according to claim 2, characterized in that The step of updating the first candidate brain region feature sequence and the second candidate brain region feature sequence based on the initial classification accuracy value to obtain high-resolution features of the first candidate brain region feature sequence and high-resolution features of the second candidate brain region feature sequence includes: Performing feature deletion on the first candidate brain region feature sequence and the second candidate brain region feature sequence to obtain an updated feature sequence set; Inputting the updated feature sequence set into the brain region state classification model to obtain an updated classification accuracy value of the updated feature sequence set; When the updated classification accuracy value is greater than the initial classification accuracy value, determining the features in the first candidate brain region feature sequence and the second candidate brain region feature sequence and not in the updated feature sequence set as low-resolution features; The low-resolution features are removed from the first candidate brain region feature sequence to obtain high-resolution features of the first candidate brain region feature sequence, and the low-resolution features are removed from the second candidate brain region feature sequence to obtain high-resolution features of the second candidate brain region feature sequence.

4. The method according to any one of claims 1 to 3, characterized in that: The target users include multiple ones; the step of obtaining the first brain region feature matrix of the target user before the image vision training and the second brain region feature matrix after the image vision training includes: Collecting a first brain image of each target user before image vision training and a second brain image of each target user after image vision training; For each target user, extracting a first feature matrix of a first brain image of the target user in multiple brain regions, and extracting a second feature matrix of a second brain image of the target user in multiple brain regions; The first feature matrices of the target users are fused to obtain the first brain region feature matrix; and the second feature matrices of the target users are fused to obtain the second brain region feature matrix.

5. The method according to claim 4, characterized in that The collecting of the first brain image of each target user before the image vision training and the second brain image of each target user after the image vision training includes: Acquire a first brain image sequence and a second brain image sequence of a plurality of candidate users; Performing head motion correction on the first brain image sequence and the second brain image sequence of each candidate user to determine the head motion range of each candidate user; Determine a candidate user whose head movement range is less than a preset movement threshold as a target user; A first brain image of each target user is determined from a first brain image sequence of each target user, and a second brain image of each target user is determined from a second brain image sequence of each target user.

6. The method according to claim 4, characterized in that The step of extracting a first feature matrix of the first brain image of the target user in multiple brain regions, and the step of extracting a second feature matrix of the second brain image of the target user in multiple brain regions, comprises: Acquire a low-frequency amplitude feature matrix, a local consistency feature matrix, and a functional connection feature matrix of the first brain image in multiple brain regions, and fuse the low-frequency amplitude feature matrix, the local consistency feature matrix, and the functional connection feature matrix to obtain the first feature matrix; A low-frequency amplitude feature matrix, a local consistency feature matrix, and a functional connectivity feature matrix of the second brain image in multiple brain regions are obtained, and the low-frequency amplitude feature matrix, the local consistency feature matrix, and the functional connectivity feature matrix are fused to obtain the second feature matrix.

7. The method according to any one of claims 1 to 3, characterized in that: The determining, according to the feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix, a first candidate brain region feature sequence of the target user before the image vision training and a second candidate brain region feature sequence after the image vision training, comprises: Calculating a quantified value of feature difference between the first brain region feature matrix and the second brain region feature matrix; For a feature difference quantization value of any brain region, if the feature difference quantization value is greater than a preset difference quantization threshold, determining a brain region identifier corresponding to the feature difference quantization value; The brain region features matching the brain region identifier in the first brain region feature matrix are determined as a first candidate brain region feature sequence of the target user, and the brain region features matching the brain region identifier in the second brain region feature matrix are determined as a second candidate brain region feature sequence of the target user.

8. A brain feature recognition device, characterized in that: The device comprises: A data acquisition module, used to acquire a first brain region feature matrix of a target user before image vision training and a second brain region feature matrix after image vision training; a feature screening module, for determining a first candidate brain region feature sequence of the target user before image vision training and a second candidate brain region feature sequence after image vision training according to a feature difference quantization value between the first brain region feature matrix and the second brain region feature matrix; The feature recognition module is used to determine the brain change characteristics of the target user during the image vision training process according to the first candidate brain region feature sequence and the second candidate brain region feature sequence.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images