Mental disease diagnosis method and system based on neural image cross-scale fusion method

By adopting a cross-scale fusion method in the diagnosis of mental illness, voxel-level indicators of multimodal magnetic resonance images are extracted and frequency-domain analysis is carried out, the problem of insufficient utilization of modal information in the prior art is solved, and the diagnostic accuracy is improved. Reduced information loss.

CN119067922AActive Publication Date: 2024-12-03SHANDONG JIANZHU UNIV

Patent Information

Application Number
CN202411079479.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-03
Estimated Expiration
2044-08-07

AI Technical Summary

Technical Problem

The prior art ignores covariant information and complementary information between different modal data in the diagnosis of mental illness, and usually only single representative indicators are extracted for fusion analysis, resulting in low accuracy of the diagnostic model and unable to provide more accurate suggestions.

Method used

Using a cross-scale fusion method based on neural image, the multimodal magnetic resonance image is obtained for preprocessing, voxel-level index values ​​are extracted, standardized and discrete Fourier transforms, frequency domain features are extracted, and the trained diagnostic model is used. Produce mental illness type prediction.

Benefits of technology

Effectively explore potential information within and between modals, improve the accuracy of model diagnosis, and reduce information loss through fine-grained calculations and new indicator algorithms describing frequency domain data, providing more comprehensive and detailed diagnosis result.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119067922A_ABST
    Figure CN119067922A_ABST
Patent Text Reader

Abstract

The invention belongs to the field of brain images and the field of deep learning, and provides a mental disease diagnosis method and system based on a neural image cross-scale fusion method. The mental disease diagnosis method based on the neural image cross-scale fusion method comprises the following steps: acquiring a tested brain multi-modal magnetic resonance image and carrying out image preprocessing to obtain a brain multi-modal magnetic resonance image with uniform voxel size; extracting an index value of each voxel of the subject from the brain multi-modal magnetic resonance image with the uniform voxel size to form a subject index matrix; independently normalizing each row of indexes in the tested index matrix to obtain a standardized index matrix; fusing data of each column in the standardized index matrix by using a discrete Fourier transform algorithm to obtain frequency domain information, and extracting features of corresponding voxels by using a frequency domain distribution index; and processing the features of each voxel by using the trained diagnosis model, predicting the probability of the mental disease types, and outputting the mental disease type with the maximum probability.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of brain imaging and deep learning, and in particular relates to a method and system for diagnosing mental illness based on a cross-scale fusion method of neuroimaging. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] At present, the diagnosis of mental illness based on magnetic resonance imaging mainly focuses on improving the accuracy of the diagnostic model by improving the learning model and introducing new optimization methods in single-modal data, ignoring the covariation information and complementary information between different modal data. In addition, in existing multimodal fusion research, usually only one representative indicator is extracted from each modality for fusion analysis or feature extraction with other modal indicators, and each modal data is limited to a single indicator for description. This may result in the data information still having potential for mining and the accuracy of the diagnostic model being reduced, and it is impossible to provide doctors with more accurate advice. Summary of the invention

[0004] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides a method and system for diagnosing mental illness based on a cross-scale fusion method of neural imaging, which can more fully mine and utilize the characteristic information in brain magnetic resonance imaging data, and more comprehensively explore the abnormalities of brain functional connections and their pathogenesis in patients with mental illness.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] The first aspect of the present invention provides a method for diagnosing mental illness based on a cross-scale fusion method of neuroimaging.

[0007] A method for diagnosing mental illness based on a cross-scale fusion method of neuroimaging, comprising:

[0008] Acquire the subject's brain multimodal magnetic resonance images and perform image preprocessing to obtain the brain multimodal magnetic resonance images with uniform voxel size;

[0009] Extract the index value of each voxel of the subject from the multimodal brain magnetic resonance images with uniform voxel size to form the index matrix of the subject;

[0010] Each row of indicators in the subject's indicator matrix is ​​normalized separately to obtain a standardized indicator matrix;

[0011] The discrete Fourier transform algorithm is used to fuse the data of each column in the standardized indicator matrix to obtain frequency domain information, and then the frequency domain distribution index is used to extract the characteristics of the corresponding voxels;

[0012] The trained diagnostic model is used to process the features of each voxel, predict the probability of the mental illness type and output the mental illness type with the highest probability.

[0013] The second aspect of the present invention provides a mental illness diagnosis system based on a cross-scale fusion method of neuroimaging.

[0014] A mental illness diagnosis system based on a cross-scale fusion method of neuroimaging, comprising:

[0015] A multimodal magnetic resonance image preprocessing module, which is used to obtain a multimodal magnetic resonance image of the subject's brain and perform image preprocessing to obtain a multimodal magnetic resonance image of the brain with a uniform voxel size;

[0016] An indicator matrix construction module is used to extract the indicator value of each voxel of the subject from the multimodal brain magnetic resonance images with uniform voxel size to form the indicator matrix of the subject;

[0017] An indicator standardization module is used to normalize each row of indicators in the indicator matrix of the subject separately to obtain a standardized indicator matrix;

[0018] A feature extraction and fusion module is used to fuse the data of each column in the standardized indicator matrix using a discrete Fourier transform algorithm to obtain frequency domain information, and then use the frequency domain distribution index to extract the features of the corresponding voxels;

[0019] The mental illness type prediction module is used to process the characteristics of each voxel using the trained diagnostic model, predict the probability of the mental illness type and output the mental illness type with the highest probability.

[0020] A third aspect of the present invention provides a computer-readable storage medium.

[0021] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps in the method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging as described above.

[0022] A fourth aspect of the present invention provides a computer device.

[0023] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging as described above are implemented.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] (1) The present invention takes into account the deficiency of using only one representative feature for each modality in multimodal fusion of brain magnetic resonance imaging. Based on the cross-scale idea, multiple computable indicators are extracted and then modal fusion is performed, which effectively mines the potential information within and between modalities, and helps to improve the accuracy of model diagnosis.

[0026] (2) Compared with the current coarse-grained method based on the region of interest (ROI), the present invention uses fine-grained calculation indicators and fusion modalities based on the voxel level, which can minimize the loss of information in the initial data preparation work, and the retained information will be more comprehensive and detailed. Finally, in order to better extract the fused features, we also proposed a new indicator algorithm to describe discrete frequency domain data. Compared with other existing commonly used indicators, this indicator does not quantify the frequency domain data from a single aspect, but comprehensively describes and refines the frequency range distribution and frequency fluctuation degree, further ensuring that the loss of data information is reduced.

[0027] Advantages of additional aspects of the present invention will be given in part in the following description, and in part will become obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0029] Figure 1 This is a flow chart of a method for diagnosing mental illness based on a cross-scale fusion method of neuroimaging according to an embodiment of the present invention.

[0030] Figure 2 It is a schematic diagram of constructing a subject indicator matrix according to an embodiment of the present invention.

[0031] Figure 3 2 is a schematic diagram of “extracting high amplitude frequency band interval position coefficients” according to an embodiment of the present invention.

[0032] Figure 4 It is a schematic diagram of peak grouping involved in extracting peak distribution coefficients according to an embodiment of the present invention.

[0033] Figure 5 It is a schematic diagram comparing different maximum window parameters involved in extracting peak distribution coefficients in an embodiment of the present invention.

[0034] Figure 6 It is a schematic diagram of the maximum window algorithm iteration involved in extracting the peak distribution coefficient in an embodiment of the present invention.

[0035] Figure 7 2 is a schematic diagram comparing different fault tolerance parameters involved in “extracting peak distribution coefficient” in an embodiment of the present invention.

[0036] Figure 8 It is a schematic diagram comparing different frequency domain results involved in extracting the global amplitude fluctuation coefficient according to an embodiment of the present invention.

[0037] Fig. 9 It is a schematic diagram of extracting the global amplitude fluctuation coefficient according to an embodiment of the present invention.

[0038] Fig.10 It is a schematic diagram of training a neural network diagnosis model in step S105 of an embodiment of the present invention.

[0039] Fig.11 It is a schematic diagram of the structure of a mental illness diagnosis system based on a cross-scale fusion method of neuroimaging according to an embodiment of the present invention. DETAILED DESCRIPTION

[0040] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0041] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0042] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0043] Terminology explanation:

[0044] A voxel is the basic unit in an fMRI image, representing a volume element in three-dimensional space being measured. Each voxel contains information about the brain region at its location. By collecting data from many voxels in the brain, a detailed picture of brain activity can be created.

[0045] Embodiment 1

[0046] like Figure 1 As shown, this embodiment provides a method for diagnosing mental illness based on a neuroimaging cross-scale fusion method, which includes:

[0047] S101: Acquire the subject's brain multimodal magnetic resonance image and perform image preprocessing to obtain the brain multimodal magnetic resonance image with uniform voxel size;

[0048] S102: extracting the index value of each voxel of the subject from the multimodal brain magnetic resonance image with uniform voxel size to form an index matrix of the subject;

[0049] S103: normalize each row of indicators in the indicator matrix of the subject separately to obtain a standardized indicator matrix;

[0050] S104: using a discrete Fourier transform algorithm to fuse the data of each column in the standardized indicator matrix to obtain frequency domain information, and then using the frequency domain distribution index to extract the features of the corresponding voxels;

[0051] S105: Use the trained diagnostic model to process the features of each voxel, predict the probability of the type of mental illness and output the type of mental illness with the highest probability.

[0052] In step S101, the subjects are from international public databases or databases obtained by self-scanning. The international public databases include Autism Imaging Data Exchange (ABIDE) and Alzheimer's Disease National Institute (ADNI), and structural magnetic resonance images (sMRI), functional magnetic resonance images (fMRI), and diffusion magnetic resonance images (DTI) of all subjects are obtained respectively.

[0053] The preprocessing process of sMRI images includes spatial normalization, tissue segmentation, modulation and smoothing denoising.

[0054] The preprocessing process of fMRI images includes: removing time points, head motion and time layer correction, spatial normalization and smoothing filtering.

[0055] The preprocessing process of DTI images includes: head motion and eddy current correction, gradient direction correction, and tensor model fitting.

[0056] Since different modal data are based on different voxel sizes, in order to facilitate the correspondence of voxels in different indicators in the later stage, all preprocessed data are upsampled or downsampled to unify the voxel size. Assuming that there are m subjects The brain MRI data of the subjects were preprocessed, and the number of voxels in each modality was c after the voxel size was unified. voxel .

[0057] In step S102:

[0058] The index value of each voxel is extracted from the obtained preprocessing results based on different applicable algorithms. Each modality data has multiple indicators to describe the data information from multiple angles.

[0059] For example, in sMRI, the volume of each voxel can be calculated;

[0060] In fMRI, the amplitude of low-frequency fluctuations (ALFF) or the fractional amplitude of low-frequency fluctuations (fALFF) can be calculated to quantify the intensity of brain activity in the area, and the local consistency (ReHo) can be calculated to reflect the synchronization of brain functional activity states in local areas of whole-brain voxels.

[0061] The fractional anisotropy index (FA) calculated in DTI can represent the degree of anisotropic diffusion of tissue.

[0062] like Figure 2 As shown in the figure, assume that in this step, each voxel of each subject calculates n indicators. Therefore, each subject can obtain n voxel-level indicator brain maps, and then use a 6mm Gaussian kernel to smooth each brain map. After processing, each brain map is arranged according to voxels and expanded into a 1×c voxel Finally, all the one-dimensional row vectors of each subject are stacked into an n×c voxel The indicator matrix is ​​m subjects indivual.

[0063] It should be noted that in order to ensure that the data at the same position in each matrix represents the same meaning, the row order and column order of each indicator matrix are exactly the same.

[0064] In step S103:

[0065] In order to eliminate the dimensional influence between different indicators, all indicator matrices obtained by the standardization algorithm are "row standardized". Specifically. Each indicator matrix first calculates the mean μ and standard deviation σ of each row, and then calculates the new value of the row one by one according to the standardization formula. Assume that the mean of the first row of the first indicator matrix m1 calculated according to formulas (1) and (2) is μ1 and the standard deviation is σ1. Then, according to formula (3), the standardized values ​​of all voxels in this row are calculated and replaced with the old values. Repeat this step for each row n times to obtain a new indicator matrix M1, denoted as New_M i .

[0066]

[0067]

[0068]

[0069] x i is the original value before normalization. is the new value after normalization.

[0070] In step S104:

[0071] For the obtained m subjects n×cvoxel The standardized indicator matrix is ​​used to perform indicator fusion and feature extraction. Specifically, assuming that the standardized indicator matrix of the first subject is recorded as New_M1, each column is a different indicator of the unified voxel and the dimensional influence between the indicators has been eliminated.

[0072] Take the data in the first column and use the discrete Fourier transform algorithm to fuse the different index data of the same voxel and obtain the frequency domain data, denoted as f1. At this time, f1 is composed of g1 discrete points in the frequency domain.

[0073] Perform discrete Fourier transform on each column of data, and each subject will obtain c voxel Frequency domain data Then the frequency domain distribution index algorithm is used to calculate the frequency domain distribution index of each voxel Construct a new fusion feature l1.

[0074] The specific frequency domain distribution index algorithm is as follows:

[0075] Assume that the frequency domain data f1 of voxel 1 is composed of g1 discrete points (x, y) in the frequency domain, the x coordinate of each point represents the frequency, and the y coordinate represents the amplitude of the frequency. Based on the proposed new index λ1 (Formula 4), the high amplitude frequency band interval position of the frequency domain data f1 (whether the high amplitude frequency is concentrated in high frequency or low frequency, denoted as ε1), the amplitude aggregation (whether the high amplitude is aggregated or discrete, denoted as ) and relative amplitude intensity (whether the amplitude intensity is relatively gentle or changes dramatically, recorded as ρ1). The larger λ1 is, the greater the amplitude fluctuation in the frequency domain data f1 is and the more concentrated it is at low frequencies, and vice versa. The detailed formula explanation is as follows.

[0076] like Figure 4 As shown in the figure, the grouping result after sorting the peak points determines whether it is a high amplitude. In layman's terms, the amplitude value that is significantly larger than other amplitude values ​​on the same spectrum is a high amplitude, and their position on the spectrum graph is higher than other points.

[0077]

[0078] like Figure 3 As shown, the frequency α is taken as the dividing point of x, the frequency interval less than α is recorded as the low-frequency area, and the frequency interval greater than α is recorded as the high-frequency area.

[0079] The frequency values ​​with non-zero amplitudes in the spectrum graphs of all voxels are counted to obtain a set of frequency values, and then the median of the frequency value set is used as the dividing point between high frequency and low frequency, that is, frequency α.

[0080] Calculate the sum of the low-frequency amplitudes of f1 low and the sum of the high frequency amplitude SUMhigh , then calculate SUM low with SUM high If the sign is positive, it means that the total amplitude of low frequency in f1 is greater than the total amplitude of high frequency, and the amplitude fluctuation is mostly concentrated in the low frequency band. Otherwise, it means that the amplitude fluctuation of f1 is mostly concentrated in the high frequency band. The specific formula is shown in formula (4).

[0081]

[0082] There are many points in the frequency domain data f whose amplitudes are much larger than the rest of the amplitudes. Here, they are called frequency domain peak points. They are also the main reference points for distinguishing different frequency domain data. The distribution of frequency domain peak points in different frequency domain data is also inconsistent. For example, Figure 5 As shown in Figure 1, the frequency domain peak points on the left are obviously more concentrated than the two on the right in frequency distribution. In order to describe and quantify this phenomenon, a new indicator based on the maximum window length algorithm is proposed.

[0083] The specific algorithm is as follows:

[0084] Figure 4 As shown, firstly, all points of f1 are sorted in ascending order according to the amplitude to obtain the ordered y. At this time, the x of the point represents the sorting rank, and y represents the amplitude.

[0085] The second step is to start from the maximum point and loop through the slopes between the current point and the next point to obtain all the slopes. The purpose of this step is to find a trend of a steep drop in amplitude to distinguish the peak point in the frequency domain from the remaining points. Therefore, a slope threshold β is set. When the first slope is greater than β, it proves that the amplitude difference between the two points is huge. All amplitudes before the slope are classified into the high amplitude group G. big , the rest are classified as low amplitude group G small If no slope is greater than the threshold, all amplitudes are defined as high amplitude group G. big It is stipulated that as long as the inclination angle is greater than 45°, it is the boundary condition for high and low amplitudes, that is, the slope threshold β is 1.

[0086] Step 3. Figure 6 As shown, a window length parameter w and a counter C are initialized, and the sliding operation with a step length of w is repeated from low frequency to high frequency until all data points are in the window. After each sliding, the grouping combination of all points in the current window is counted according to the amplitude grouping in the second step. When there are two or more points in the window that are in the high amplitude group G bigWhen , the counter C is incremented by 1. After the window slide is completed, the counter C value is compared with the fault tolerance rate γ. When c is greater than or equal to γ, it means that the current window length is too large for f1, and the w parameter is reduced and the third step is repeated; if c is less than γ, it proves that the current window length satisfies f1, the algorithm stops and outputs the maximum window length w. Calculate the index of f1 according to formula (5)

[0087]

[0088] w1 represents the maximum window length value of the first voxel; the iteration stops as long as 80% of the window structures meet the condition. Therefore, the error tolerance γ is 20%.

[0089] Figure 5 As shown, The smaller it is, the fewer and more dispersed the peak points in the frequency domain are. The larger the value, the more and more concentrated the peak points in the frequency domain are. Figure 7 As shown, the smaller the error tolerance rate γ is, the stricter the maximum window algorithm is.

[0090] Relative amplitude intensity refers to whether the amplitude intensity is relatively flat or changes dramatically. It is mainly intended to quantify and describe the difference in amplitude between different frequency domain data f. Figure 8 As shown, even though they have similar shapes, their vibration intensities are obviously different, so the global amplitude fluctuation coefficient ρ is used to measure them.

[0091] The specific algorithm steps are as follows:

[0092] like Fig. 9 As shown, first, the y in the discrete Fourier transform results of all voxels are sorted in ascending order to obtain a sorting ranking diagram and a distribution histogram. In the sorting ranking diagram, x represents the sorting ranking, and y represents the amplitude size. The larger the x, the larger the amplitude. In the distribution histogram, x represents the amplitude size, and y represents the number of times the amplitude size appears. The larger the y, the more frequently the amplitude size appears.

[0093] The second step is to calculate the weighted ranking sum of all amplitudes y in a voxel according to formula (6):

[0094]

[0095] Represents the rank of the i-th amplitude value y in the amplitude sorting table;

[0096] in Represents the number of times the amplitude y appears. This shows that the more times an amplitude appears, the smaller its corresponding weight is, and the less times an amplitude appears, the larger its corresponding weight is. The ρ1 calculated in this way will widen the gap between the frequency domain data f and increase the recognition of each frequency domain data f. Figure 8 As shown in the figure, the frequency domain data f with a greater amplitude fluctuation will be more dispersed in the sorting order diagram, and the corresponding ρ1 will be larger, and vice versa. Therefore, p can well describe the relative amplitude strength information of each f.

[0097] Perform the above steps on all column vectors in the standardized indicator matrix of the subject to obtain a 1×c voxel The fusion feature l1.

[0098] Performing the above operation on the standardized indicator matrix of all subjects can obtain m subjects 1×c voxel Fusion features

[0099] In step S105: the training process of the diagnostic model is:

[0100] Using the fusion features extracted above The dataset is used to train a deep learning model to generate the final diagnosis results.

[0101] like Fig.10 As shown, first, all the fusion features are stacked into an m subjects ×c voxel The characteristic matrix of . Since the dimension of this matrix is ​​too high, the commonly used algorithms such as principal component analysis (PCA) can be used to reduce the dimension of the matrix.

[0102] Then, the reduced feature matrix is ​​input into the fully connected neural network. The fully connected neural network is a binary classification network consisting of an input layer, an intermediate layer, and an output layer. During the training process, a 5-fold cross-validation strategy is used to divide the training set into 5 parts, 4 of which are used as training sets and the other 1 as a test set to obtain 5 classification accuracy results, and then the average of the 5 results is used as the final evaluation indicator of this experiment.

[0103] Continue to iterate and adjust the parameters until the model results reach the optimal state and then stop the training step.

[0104] The present invention proposes a mental illness diagnosis system based on a cross-scale fusion method of neuroimaging. The present invention uses multimodal magnetic resonance brain imaging data, and from the perspective of mining potential information across scales at a fine-grained level, integrates methods such as brain function index extraction, discrete Fourier transform, and deep learning models. With the help of commonly used brain magnetic resonance imaging data description methods, different voxel-level index data are extracted for each modality data, and the index data of each voxel is discrete Fourier transformed for fusion and feature extraction based on the coordinate positioning method. Finally, the extracted features are used to train a deep learning model with an attention mechanism for classification diagnosis. By comparing the non-cross-scale method with the method using coarse-grained features, it can be seen that this method can maximize the comprehensive coverage of data information and improve classification accuracy.

[0105] Embodiment 2

[0106] like Fig.11 As shown, this embodiment provides a mental illness diagnosis system based on a neuroimage cross-scale fusion method, which includes:

[0107] A multimodal magnetic resonance image preprocessing module, which is used to obtain a multimodal magnetic resonance image of the subject's brain and perform image preprocessing to obtain a multimodal magnetic resonance image of the brain with a uniform voxel size;

[0108] An indicator matrix construction module is used to extract the indicator value of each voxel of the subject from the multimodal brain magnetic resonance images with uniform voxel size to form the indicator matrix of the subject;

[0109] An indicator standardization module is used to normalize each row of indicators in the indicator matrix of the subject separately to obtain a standardized indicator matrix;

[0110] A feature extraction and fusion module is used to fuse the data of each column in the standardized indicator matrix using a discrete Fourier transform algorithm to obtain frequency domain information, and then use the frequency domain distribution index to extract the features of the corresponding voxels;

[0111] The mental illness type prediction module is used to process the characteristics of each voxel using the trained diagnostic model, predict the probability of the mental illness type and output the mental illness type with the highest probability.

[0112] It should be noted here that the various modules in the mental illness diagnosis system based on the cross-scale fusion method of neural images correspond one to one with the various steps in the above-mentioned mental illness diagnosis method based on the cross-scale fusion method of neural images, and the specific implementation process is the same, which will not be repeated here.

[0113] Embodiment 3

[0114] This embodiment uses a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the steps in the method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging as described above are implemented.

[0115] Embodiment 4

[0116] This embodiment provides a computer device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the steps in the method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging as described above are implemented.

[0117] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products of the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of the processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for diagnosing mental illness based on a cross-scale fusion method of neural images, characterized in that: include: Acquiring multimodal brain magnetic resonance images of the subjects and performing image preprocessing to obtain multimodal brain magnetic resonance images with uniform voxel size; obtaining multimodal brain magnetic resonance images with uniform voxel size; Extract the index value of each voxel of the subject from the multimodal brain magnetic resonance images with uniform voxel size to form the index matrix of the subject; Each row of indicators in the subject's indicator matrix is ​​normalized separately to obtain a standardized indicator matrix; The discrete Fourier transform algorithm is used to fuse the data of each column in the standardized indicator matrix to obtain frequency domain information, and then the frequency domain distribution index is used to extract the characteristics of the corresponding voxels; The trained diagnostic model is used to process the features of each voxel, predict the probability of the mental illness type and output the mental illness type with the highest probability.

2. The method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging according to claim 1, characterized in that: The process of extracting the index value of each voxel of the subject from the results of image preprocessing and forming the index matrix of the subject is as follows: According to the index of each voxel of each subject, the voxel-level index brain map corresponding to each subject is obtained; Each brain map was then smoothed using a Gaussian kernel; After processing, each brain map is arranged according to voxels and expanded into a one-dimensional row vector. Finally, all one-dimensional row vectors of each subject are stacked into an indicator matrix.

3. The method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging according to claim 1, characterized in that: Each row of indicators in the subject's indicator matrix is ​​normalized separately to obtain the standardized indicator matrix in the following process: first calculate the mean and standard deviation of each row of each indicator matrix, and then calculate the new value of the row one by one according to the standardized formula.

4. The method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging according to claim 1, characterized in that: The steps of using the discrete Fourier transform algorithm to fuse the data of each column in the standardized indicator matrix are: Take the data of each column of the standardized indicator matrix of each subject and use the discrete Fourier transform algorithm to fuse the different indicator data of the same voxel and obtain several frequency domain data; The frequency domain distribution index algorithm is used to calculate the frequency domain distribution index of each voxel to construct a new fusion feature; wherein, the formula for calculating the frequency domain distribution index of each voxel using the frequency domain distribution index algorithm is: the frequency domain distribution index is the product of the high amplitude frequency band interval position, amplitude aggregation and relative amplitude intensity of the current frequency domain data.

5. A mental illness diagnosis system based on a cross-scale fusion method of neural images, characterized in that: include: A multimodal magnetic resonance image preprocessing module, which is used to obtain a multimodal magnetic resonance image of the subject's brain and perform image preprocessing to obtain a multimodal magnetic resonance image of the brain with a uniform voxel size; An indicator matrix construction module is used to extract the indicator value of each voxel of the subject from the multimodal brain magnetic resonance images with uniform voxel size to form the indicator matrix of the subject; An indicator standardization module is used to normalize each row of indicators in the indicator matrix of the subject separately to obtain a standardized indicator matrix; A feature extraction and fusion module is used to fuse the data of each column in the standardized indicator matrix using a discrete Fourier transform algorithm to obtain frequency domain information, and then use the frequency domain distribution index to extract the features of the corresponding voxels; The mental illness type prediction module is used to process the characteristics of each voxel using the trained diagnostic model, predict the probability of the mental illness type and output the mental illness type with the highest probability.

6. The mental illness diagnosis system based on the cross-scale fusion method of neuroimaging according to claim 5, characterized in that: The process of extracting the index value of each voxel of the subject from the results of image preprocessing and forming the index matrix of the subject is as follows: According to the index of each voxel of each subject, the voxel-level index brain map corresponding to each subject is obtained; Each brain map was then smoothed using a Gaussian kernel; After processing, each brain map is arranged according to voxels and expanded into a one-dimensional row vector. Finally, all one-dimensional row vectors of each subject are stacked into an indicator matrix.

7. The mental illness diagnosis system based on the cross-scale fusion method of neuroimaging according to claim 5, characterized in that: Each row of indicators in the subject's indicator matrix is ​​normalized separately to obtain the standardized indicator matrix in the following process: first calculate the mean and standard deviation of each row of each indicator matrix, and then calculate the new value of the row one by one according to the standardized formula.

8. The mental illness diagnosis system based on the cross-scale fusion method of neuroimaging according to claim 5, characterized in that: The steps of using the discrete Fourier transform algorithm to fuse the data of each column in the standardized indicator matrix are: Take the data of each column of the standardized indicator matrix of each subject and use the discrete Fourier transform algorithm to fuse the different indicator data of the same voxel and obtain several frequency domain data; The frequency domain distribution index algorithm is used to calculate the frequency domain distribution index of each voxel to construct a new fusion feature; wherein, the formula for calculating the frequency domain distribution index of each voxel using the frequency domain distribution index algorithm is: the frequency domain distribution index is the product of the high amplitude frequency band interval position, amplitude aggregation and relative amplitude intensity of the current frequency domain data.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps in the method for diagnosing mental illness based on a cross-scale fusion method of neuroimaging as described in any one of claims 1 to 4 are implemented.

10. A computer device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the steps in the method for diagnosing mental illness based on the cross-scale fusion method of neuroimaging as described in any one of claims 1 to 4 are implemented.

Citation Information

Patent Citations

  • Multi-modal medical image fusion method based on deep learning

    CN115272261A

  • Multi-mode anti-fraud identity recognition method and device, equipment and storage medium

    CN117151721A

  • Multi-modal image fusion method and system specific to diagnosis group

    CN118135318A

  • Magnetic resonance data classification system

    CN118429734A

  • KR20230015706A

Cited By

  • Method for identifying autism brain map through multi-mode brain map information fusion

    CN119888325A