A method and device for generating a reference brain image, an electronic device, and a storage medium

Abstract

The application provides a method and device for generating a reference brain image, electronic equipment and a storage medium, wherein the method comprises: in response to a received evaluation instruction, obtaining an individual brain image to be evaluated indicated by the evaluation instruction; and inputting the individual brain image into a trained variational autoencoder model to obtain a reference brain image corresponding to the individual brain image output by the variational autoencoder model, the reference brain image being used to indicate brain features in a healthy state of the individual to be evaluated. The AI model is used to reconstruct the brain image of the individual to be evaluated based on brain features of a normal population, so that a more targeted reference brain image is obtained, a personalized normal reference brain image is formed, and the individual health evaluation is better assisted.

Description

Technical Field

[0001] This application relates to the field of medical image processing technology, and more specifically, to a method, apparatus, electronic device, and storage medium for generating reference brain images. Background Technology

[0002] In current clinical practice, the determination of individual brain imaging abnormalities mainly relies on universal reference standards for the normal population (such as population mean brain structure morphology, anatomical templates, etc.). However, such universal standards are difficult to take into account physiological differences caused by individual age, sex, genetic background, lifestyle habits, etc., which can easily lead to false positive diagnoses or missed diagnoses of early, minor pathological changes. Taking Alzheimer's disease as an example, clinicians compare the brain structure templates of the normal population to determine whether an individual has characteristic changes such as hippocampal atrophy. However, there are natural differences in the basic morphology of the hippocampus among different individuals. Universal standards may mistakenly identify physiological individual differences as pathological changes or overlook early pathological atrophy. Summary of the Invention

[0003] The purpose of this application is to provide a method, apparatus, electronic device and storage medium for generating reference brain images, so as to solve the technical problem that the general reference standard for brain images affects the accuracy of individual judgment results in the prior art.

[0004] In a first aspect, the present invention provides a method for generating a reference brain image, the method comprising: in response to a received evaluation instruction, acquiring an individual brain image to be evaluated as indicated by the evaluation instruction; inputting the individual brain image into a trained variational autoencoder model to obtain a reference brain image corresponding to the individual brain image output by the variational autoencoder model, the reference brain image being used to indicate brain features of the individual to be evaluated under healthy conditions.

[0005] In optional implementations, the method also includes comparing individual brain images with reference brain images to help determine the patient's brain health assessment results.

[0006] In an optional implementation, the variational autoencoder model includes at least an encoder, a vector quantization module, and a decoder. The encoder maps individual brain images to a continuous latent space. The vector quantization module quantizes the continuous latent space using a discrete codebook to output a discrete latent space feature vector of the individual brain image. Based on the feature differences between the discrete latent space feature vector of the individual brain image and the discrete latent space feature vector of a normal brain image, it determines whether to correct the discrete latent space feature vector of the individual brain image. If so, it resamples based on the standard latent space feature vector to generate a corrected latent space feature vector of the individual brain image. The decoder reconstructs the image using the corrected latent space feature vector of the individual brain image to output a reference brain image.

[0007] In an optional implementation, the latent space feature vector includes multiple codewords. The step of determining whether to modify the discrete latent space feature vector of the individual brain image based on the feature differences between the discrete latent space feature vector of the individual brain image and the discrete latent space feature vector of a normal brain image specifically includes: Statistical analysis of the deviation values ​​of codewords / codeword combinations in the discrete latent space feature vectors of individual brain images; For each codeword / codeword combination, determine whether the codeword / codeword combination needs to be corrected based on the relationship between the corresponding deviation value and the preset value; If necessary, mark the codeword / codeword combination.

[0008] In an optional implementation, the step of resampling based on the standard latent space feature vector to generate a modified latent space feature vector for an individual brain image specifically includes: For each labeled codeword / codeword combination, determine the codeword / codeword combination at the corresponding position in the standard latent space feature vector; The tagged codewords / codeword combinations are replaced with the codewords / codeword combinations at the corresponding positions in the standard latent space feature vector.

[0009] In an optional implementation, the standard latent space feature vector is determined in the following manner: Multiple normal brain images were collected to train the original variational autoencoder model. The model parameters and codebook were iteratively optimized, and the trained variational autoencoder model and the discrete latent space feature vector corresponding to each normal brain image were obtained. Random sampling was performed within the quartile intervals of the discrete latent space feature vectors corresponding to all normal brain images to serve as standard latent space feature vectors.

[0010] In an optional implementation, the preset value is determined in the following manner: The abnormal brain images are input into the trained variational autoencoder model to obtain the discrete latent space feature vectors corresponding to the abnormal brain images. Based on the discrete latent space feature vectors corresponding to abnormal brain images and the codeword / codeword combination distribution probabilities in the discrete latent space feature vectors of normal brain images, preset values ​​are determined.

[0011] Secondly, the present invention provides a reference brain image generation apparatus, the apparatus comprising: The response module is used to acquire the individual brain image to be evaluated as indicated by the received evaluation instruction. The processing module is used to input individual brain images into a trained variational autoencoder model to obtain a reference brain image corresponding to the individual brain image output by the variational autoencoder model. The reference brain image is used to indicate the brain characteristics of the individual to be evaluated under healthy conditions.

[0012] Thirdly, the present invention provides an electronic device, comprising: a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of any of the aforementioned methods for generating reference brain images.

[0013] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the steps of the method for generating a reference brain image as described in any of the foregoing embodiments.

[0014] This application provides a method, apparatus, electronic device, and storage medium for generating reference brain images. The method includes, in response to a received assessment instruction, acquiring a brain image of the individual to be assessed as indicated by the assessment instruction; inputting the individual brain image into a trained variational autoencoder model to obtain a reference brain image corresponding to the individual brain image output by the variational autoencoder model. The reference brain image is used to indicate brain characteristics of the individual to be assessed under healthy conditions. By reconstructing the brain image of the individual to be assessed based on brain characteristics of a normal population using an AI model, a more targeted reference brain image is obtained, forming a personalized, normal reference brain image, thereby better assisting in individual health assessment. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 A flowchart illustrating a method for generating a reference brain image, provided in an embodiment of this application; Figure 2 A schematic diagram of a technical route provided for an embodiment of this application; Figure 3 A schematic diagram of the structure of a reference brain image generation device provided in an embodiment of this application; Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0018] Figure 1This is a schematic flowchart illustrating a method for generating a reference brain image, provided as an embodiment of this application. Figure 1 As shown, an embodiment of this application provides a method for generating a reference brain image, comprising: S10. In response to the received assessment instruction, acquire the individual brain image to be assessed as indicated by the assessment instruction.

[0019] The assessment instructions here can be used to specify the brain images of the individual to be assessed. The brain images can be magnetic resonance imaging (MRI) of the brain of the individual to be assessed, such as T1wMRI, and are usually in NIfTI (.nii) or DICOM format.

[0020] Before processing, individual brain images can be preprocessed, such as spatial normalization, using standard brain atlases (e.g., MNI152) to register the original images, mapping the individual brain to a standard space to ensure the applicability of subsequent models. Intensity normalization can also be performed, applying linear scaling or nonlinear correction to signal intensity to eliminate differences between scanners and bias effects. Brain extraction can also be performed: algorithms such as BET and HD-BET can be used to remove non-brain tissues such as the skull and scalp, retaining only the brain parenchyma. Resolution matching can also be performed, resampling the images to the spatial resolution used to train the VQ-VAE model (e.g., 1 cubic millimeter isotropic voxel).

[0021] S11. Input the individual brain image into the trained variational autoencoder model to obtain the reference brain image corresponding to the individual brain image output by the variational autoencoder model. The reference brain image is used to indicate the brain characteristics of the individual to be evaluated under healthy conditions.

[0022] In step S2, a model can be constructed to predict a reference brain image of the individual under healthy conditions based on the individual brain image of the individual to be evaluated.

[0023] Specifically, the variational autoencoder model includes at least an encoder, a vector quantization module, and a decoder. The encoder maps individual brain images to a continuous latent space. The vector quantization module quantizes the continuous latent space using a discrete codebook to output a discrete latent space feature vector for the individual brain image. Based on the feature differences between the discrete latent space feature vector of the individual brain image and that of a normal brain image, the vector quantization module determines whether to correct the discrete latent space feature vector of the individual brain image. If so, it resamples based on the standard latent space feature vector to generate a corrected latent space feature vector for the individual brain image. The decoder reconstructs the image using the corrected latent space feature vector of the individual brain image to output a reference brain image.

[0024] The latent space feature vector here includes multiple codewords. The step of determining whether to modify the discrete latent space feature vector of the individual brain image based on the feature differences between the discrete latent space feature vector of an individual brain image and the discrete latent space feature vector of a normal brain image can specifically include: Statistical analysis of the deviation values ​​of codewords / codeword combinations is performed on the discrete latent space feature vectors of individual brain images. For each codeword / codeword combination, based on the relationship between the corresponding deviation value and the preset value, it is determined whether the codeword / codeword combination needs to be corrected. If so, the codeword / codeword combination is marked.

[0025] For example, the distribution of each codeword or codeword combination can be statistically analyzed, such as frequency of occurrence and spatial co-occurrence patterns, and the deviation value of each codeword or codeword combination can be calculated (such as KL divergence, t-statistic, and effect size Cohen's d). The codewords or codeword combinations here can be determined based on the disease.

[0026] For example, if the deviation value of a codeword or codeword combination is less than a preset value, it is identified as a slight deviation dimension, judged as an individual physiological difference, and retained. If the deviation value of a codeword or codeword combination is greater than or equal to the preset value, it indicates a significant deviation in the corresponding disease group, is judged as a potential pathologically related feature, and marked as a dimension that needs correction.

[0027] Here, the preset value can be determined in the following way: Abnormal brain images are input into a trained variational autoencoder model to obtain discrete latent space feature vectors corresponding to the abnormal brain images. Based on the codeword / codeword combination distribution probabilities in the discrete latent space feature vectors of abnormal and normal brain images, preset values ​​are determined.

[0028] Specifically, the steps of resampling based on the standard latent space feature vector to generate a modified latent space feature vector for an individual brain image may include: For each labeled codeword / codeword combination, determine the corresponding codeword / codeword combination at the position in the standard latent space feature vector. Replace the labeled codeword / codeword combination with the corresponding codeword / codeword combination at the position in the standard latent space feature vector.

[0029] Furthermore, the standard latent space feature vector can be determined in the following way: Multiple normal brain images were acquired to train the original variational autoencoder model. The model parameters and codebook were iteratively optimized, and the trained variational autoencoder model and the discrete latent space feature vector corresponding to each normal brain image were obtained. Random sampling was performed within the quartile interval of the discrete latent space feature vectors corresponding to all normal brain images to serve as the standard latent space feature vector.

[0030] In this embodiment, selective resetting can be performed based on the characteristic distribution of the normal population. For codewords or codeword combinations that need correction, the original codeword index at the current location is obtained. The codeword distribution of the normal population at this spatial location is then consulted, i.e., the probability distribution of codewords used by healthy individuals at the same anatomical location. Weighted random sampling is performed within the quartile range of the normal population (i.e., the 25%-75% quartile range of the probability distribution). Limiting the sampling to the middle 50% range is to avoid introducing extremely rare codewords and ensure the typicality and reasonableness of the reconstruction results.

[0031] The prediction model here can also be trained and generated using other models such as GAN or Diffusion; no specific limitation is made here.

[0032] S12. Compare individual brain images with reference brain images to help determine the patient's brain health assessment results.

[0033] Taking Alzheimer's disease as an example, clinical practice involves comparing brain structure templates of normal individuals to determine whether an individual has characteristic changes such as hippocampal atrophy. However, there are natural differences in the basic morphology of the hippocampus among different individuals. The general standard may mistakenly identify physiological individual differences as pathological changes or ignore early pathological atrophy.

[0034] In this embodiment, compared with the existing technology that mainly relies on general reference standards of normal populations (such as the population mean of brain structure morphology, anatomical templates, etc.) to determine individual brain imaging abnormalities in clinical practice, this application can generate more personalized reference brain images for the individual to be evaluated. This can reduce the probability of false positive diagnosis or missed diagnosis of early minor pathological changes due to physiological differences between individuals such as age, gender, genetic background, and lifestyle. It has important clinical significance for improving diagnostic accuracy, optimizing treatment decisions and improving patient prognosis.

[0035] Figure 2 This is a schematic diagram of a technical route provided for an embodiment of this application. For example... Figure 2 As shown in a specific embodiment of this application, a method for generating "personalized normal brain images" is provided to achieve accurate personalized diagnosis and treatment and brain health assessment, including the following steps: Step 1: Feature Extraction from Normal Brain Images and Construction of Discrete Coding Model Based on a large-scale dataset of 30,000 high-resolution structural magnetic resonance imaging (MRI) images from a multi-center nationwide population covering the entire life cycle, an image encoding-decoding model was constructed using a Vector Quantized Variational Autoencoder (VQ-VAE). This model maps the original image to a continuous latent space through the encoder, then quantizes the continuous latent space using discrete encoding vectors to form discretized latent space features. The decoder reconstructs the image based on these quantized latent space features. During training, the discrete encoding vector parameters are iteratively optimized to determine a stable codebook, establishing individual-level discrete latent space representations and the overall distribution pattern of the discrete latent space in the normal population. Discrete encoding vector parameters... Step 2: Encoding abnormal brain imaging features and quantifying the differences from normal distribution The trained VQ-VAE encoder weight parameters and codebook are fixed and applied to brain image datasets of various neurological diseases (including space-occupying lesions and non-space-occupying lesions). This completes the encoding mapping and discrete latent space feature extraction of brain images for each disease group, constructing discrete latent space feature distribution models for various diseases. Through statistical tests and feature difference analysis, the degree of deviation and specific deviation dimension of the latent space features of each disease (space-occupying / non-space-occupying) from the distribution pattern of the normal population are quantified. This provides a quantitative basis for assigning higher sampling selection probabilities to features with high deviations during subsequent recoding.

[0036] Step 3: Targeted coding correction based on distribution differences and generation of individualized normal brain images For the input individualized brain images, the encoding mapping is first completed through a pre-trained VQ-VAE encoder to obtain its discrete latent space feature vector. Based on the disease-normal distribution difference model established in the second step, the deviation dimension and degree of the individual's latent space features from the normal population distribution are analyzed. Feature dimensions with small differences from the normal population are identified as individual-specific differences and retained. Feature dimensions with significant deviations from both the disease group level and the individual level (identified as disease-related abnormal features) are resampled and quantized within the quartile range (25%-75%) of the normal population distribution to form an optimized individualized discrete latent space. Finally, the optimized latent space features are decoded by the VQ-VAE decoder to generate an individualized normal brain image reference template.

[0037] This application constructs a "personalized normal brain image" reference template capable of direct and accurate comparison with abnormal brain images. Discretizing the feature space makes it easier to obtain the distribution pattern of the feature space. Based on the characteristics of the normal population distribution, the template is reset, focusing on disease-related features while maintaining the individual's main brain morphology and signal distribution patterns, truly forming a "personalized normal reference image." This application achieves multi-dimensional and comprehensive comparative analysis from signal intensity, local structural features to overall spatial distribution patterns and inter-regional dependencies, meeting a key requirement for overcoming existing technological bottlenecks and improving the accuracy of brain disease diagnosis. Compared to current physician image interpretation, this comparison is more intuitive and has the ability to detect subtle differences, improving the detection capability of abnormal brain images.

[0038] Figure 3 This is a schematic diagram of a device for generating reference brain images, provided as an embodiment of this application. Figure 3 As shown, based on the same inventive concept, this application also provides a reference brain image generation device 30, the device comprising: The response module 310 is used to acquire the individual brain image to be evaluated as indicated by the received evaluation instruction in response to the evaluation instruction. The processing module 320 is used to input individual brain images into a trained variational autoencoder model to obtain a reference brain image corresponding to the individual brain image output by the variational autoencoder model. The reference brain image is used to indicate the brain characteristics of the individual to be evaluated under healthy conditions.

[0039] In a preferred embodiment, the processing module 320 is further configured to compare individual brain images with reference brain images to help determine the patient's brain health assessment results.

[0040] In a preferred embodiment, the variational autoencoder model includes at least an encoder, a vector quantization module, and a decoder. The encoder maps individual brain images to a continuous latent space. The vector quantization module quantizes the continuous latent space using a discrete codebook to output a discrete latent space feature vector of the individual brain image. Based on the feature differences between the discrete latent space feature vector of the individual brain image and the discrete latent space feature vector of a normal brain image, it determines whether to correct the discrete latent space feature vector of the individual brain image. If so, it resamples based on the standard latent space feature vector to generate a corrected latent space feature vector of the individual brain image. The decoder reconstructs the image using the corrected latent space feature vector of the individual brain image to output a reference brain image.

[0041] In a preferred embodiment, the latent space feature vector includes multiple codewords. The step of determining whether to modify the discrete latent space feature vector of the individual brain image based on the feature differences between the discrete latent space feature vector of the individual brain image and the discrete latent space feature vector of a normal brain image specifically includes: Statistical analysis of the deviation values ​​of codewords / codeword combinations in the discrete latent space feature vectors of individual brain images; For each codeword / codeword combination, determine whether the codeword / codeword combination needs to be corrected based on the relationship between the corresponding deviation value and the preset value; If necessary, mark the codeword / codeword combination.

[0042] In a preferred embodiment, the step of resampling based on the standard latent space feature vector to generate a modified latent space feature vector for an individual brain image specifically includes: For each labeled codeword / codeword combination, determine the codeword / codeword combination at the corresponding position in the standard latent space feature vector; The tagged codewords / codeword combinations are replaced with the codewords / codeword combinations at the corresponding positions in the standard latent space feature vector.

[0043] In a preferred embodiment, the standard latent space feature vector is determined in the following manner: Multiple normal brain images were collected to train the original variational autoencoder model. The model parameters and codebook were iteratively optimized, and the trained variational autoencoder model and the discrete latent space feature vector corresponding to each normal brain image were obtained. Random sampling was performed within the quartile intervals of the discrete latent space feature vectors corresponding to all normal brain images to serve as standard latent space feature vectors.

[0044] In a preferred embodiment, the preset value is determined in the following manner: The abnormal brain images are input into the trained variational autoencoder model to obtain the discrete latent space feature vectors corresponding to the abnormal brain images. Based on the discrete latent space feature vectors corresponding to abnormal brain images and the codeword / codeword combination distribution probabilities in the discrete latent space feature vectors of normal brain images, preset values ​​are determined.

[0045] Please see Figure 4 , Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 4 As shown, the electronic device 400 includes a processor 410, a memory 420, and a bus 430.

[0046] The memory 420 stores machine-readable instructions executable by the processor 410. When the electronic device 400 is running, the processor 410 and the memory 420 communicate via the bus 430. When the machine-readable instructions are executed by the processor 410, the steps of a method for generating a reference brain image as described in the above method embodiment can be executed. For specific implementation details, please refer to the method embodiment, which will not be repeated here.

[0047] This application also provides a computer-readable storage medium storing a computer program. When the computer program is run by a processor, it can execute the steps of a method for generating a reference brain image as described in the above method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0048] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0049] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0050] Furthermore, the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0051] Furthermore, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0052] It should be noted that if the function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] In this document, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, without necessarily requiring or implying any such actual relationship or order between these entities or operations.

[0054] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for generating a reference brain image, characterized in that, The method includes: In response to a received assessment instruction, acquire the individual brain image to be assessed as indicated by the assessment instruction; The individual brain image is input into a trained variational autoencoder model to obtain a reference brain image corresponding to the individual brain image output by the variational autoencoder model. The reference brain image is used to indicate the brain features of the individual to be evaluated under healthy conditions.

2. The method according to claim 1, characterized in that, Also includes: The individual brain images are compared with the reference brain images to help determine the patient's brain health assessment results.

3. The method according to claim 1, characterized in that, A variational autoencoder model includes at least an encoder, a vector quantization module, and a decoder. The encoder is used to map individual brain images into a continuous hidden space; The vector quantization module quantizes the continuous latent space using a discrete codebook to output the discrete latent space feature vector of an individual brain image. Based on the feature differences between the discrete latent space feature vectors of individual brain images and those of normal brain images, it is determined whether to modify the discrete latent space feature vectors of individual brain images. If so, resampling is performed based on the standard latent space feature vector to generate a modified latent space feature vector for the individual brain image; The decoder reconstructs the image from the modified latent space feature vector of the individual brain image to output a reference brain image.

4. The method according to claim 3, characterized in that, The latent space feature vector includes multiple codewords. The step of determining whether to correct the discrete latent space feature vector of the individual brain image based on the feature difference between the discrete latent space feature vector of the individual brain image and the discrete latent space feature vector of the normal brain image specifically includes: Statistical analysis of the deviation values ​​of codewords / codeword combinations in the discrete latent space feature vectors of individual brain images; For each codeword / codeword combination, determine whether the codeword / codeword combination needs to be corrected based on the relationship between the corresponding deviation value and the preset value; If necessary, mark the codeword / codeword combination.

5. The method according to claim 4, characterized in that, The step of resampling based on the standard latent space feature vector to generate the modified latent space feature vector of the individual brain image specifically includes: For each labeled codeword / codeword combination, determine the codeword / codeword combination at the corresponding position in the standard latent space feature vector; The tagged codewords / codeword combinations are replaced with the codewords / codeword combinations at the corresponding positions in the standard latent space feature vector.

6. The method according to claim 5, characterized in that, The standard latent space feature vectors are determined in the following way: Multiple normal brain images were collected to train the original variational autoencoder model. The model parameters and codebook were iteratively optimized, and the trained variational autoencoder model and the discrete latent space feature vector corresponding to each normal brain image were obtained. Random sampling was performed within the quartile intervals of the discrete latent space feature vectors corresponding to all normal brain images to serve as standard latent space feature vectors.

7. The method according to claim 6, characterized in that, The preset value is determined in the following way: The abnormal brain images are input into the trained variational autoencoder model to obtain the discrete latent space feature vectors corresponding to the abnormal brain images. Based on the discrete latent space feature vectors corresponding to abnormal brain images and the codeword / codeword combination distribution probabilities in the discrete latent space feature vectors of normal brain images, preset values ​​are determined.

8. A device for generating reference brain images, characterized in that, The device includes: A response module is used to acquire the individual brain image to be evaluated as indicated by the received evaluation instruction. The processing module is used to input the individual brain image into a trained variational autoencoder model to obtain a reference brain image corresponding to the individual brain image output by the variational autoencoder model. The reference brain image is used to indicate the brain characteristics of the individual to be evaluated under healthy conditions.

9. An electronic device, characterized in that, include: The device includes a processor, a memory, and a bus, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is in operation, the processor communicates with the memory via the bus, and the processor executes the machine-readable instructions to perform the steps of the method for generating a reference brain image as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method for generating a reference brain image as described in any one of claims 1 to 7.

Images