A method for classifying neurotransmitter subtypes of schizophrenia

Abstract

A method for classifying neurotransmitter subtypes in schizophrenia, which relates to the field of medical technology. The main steps are as follows: constructing a CTh brain structure development norm by using publicly available and previously collected healthy control brain structure MRI data; extracting the cortical thickness CTh values of each brain region of the patient; calculating the CTh deviation of each brain region of the patient; collecting PET molecular imaging maps of neurotransmitter receptors or transporters, calculating the correlation coefficient ρ and forming a neurotransmitter spectrum related to CTh abnormality; performing dimensionality reduction on the neurotransmitter spectrum related to CTh abnormality; using the consensus clustering method to establish a final classification model for neurotransmitter subtypes in schizophrenia; inputting the data of the patient to be classified into the classification model to obtain the subtype type. The present invention indirectly reflects the possible abnormalities of the neurotransmitter system of individual SCZ patients by associating publicly available PET neurotransmitter receptor or transporter maps with individual brain structure deviations, and identifies neurotransmitter-related SCZ biological subtypes.

Description

Technical Field

[0001] The present invention relates to the field of medical technology, and particularly to a method for classifying neurotransmitter subtypes of schizophrenia. Background Art

[0002] Schizophrenia (SCZ) is a severe mental illness that commonly occurs in adolescence and early adulthood and seriously affects an individual's thinking, emotion, cognition, and behavior. Previous case-control design studies using Magnetic Resonance Imaging (MRI) technology have found extensive structural abnormalities in the brains of SCZ patients. However, the group-level comparison methods used in these studies assume that SCZ patients are a homogeneous population, and the results obtained are inconsistent, suggesting that the SCZ patient population based on the same symptomatic diagnostic criteria may consist of multiple subtypes with different neuropathological mechanisms. This biological heterogeneity can be traced back to the heterogeneity of individual neuropathology and genetic risk in SCZ patients, which hinders the development of reliable and effective objective biomarkers related to the diagnosis and treatment of SCZ. Isolating the biological subtypes of SCZ can effectively overcome heterogeneity, promote accurate exploration of the genetic and neuropathological mechanisms of SCZ, and help develop stable and reliable objective biomarkers.

[0003] The heterogeneity of SCZ is also reflected in treatment response: less than half of SCZ patients can achieve clinical cure after the first antipsychotic treatment, and about 30% of SCZ patients develop treatment-resistant SCZ after ineffective treatment with multiple antipsychotics. Antipsychotics targeting the neurotransmitter system are currently the main treatment method for SCZ. Existing drugs mainly act on dopamine receptor D2 and serotonin receptor 5-HT2a. The differences in drug treatment response indicate that there is also heterogeneity in the neurotransmitter system abnormalities in SCZ. Undoubtedly, the detection of abnormalities in the above neurotransmitter systems in the brain can guide drug selection. However, the current molecular imaging detection methods for neurotransmitter systems based on Positron Emission Tomography (PET) have the disadvantages of invasive operation, risk of drug allergy, and high economic burden. In addition, a single PET imaging can only detect one neurotransmitter receptor / transporter, but existing evidence shows that in addition to the dopamine and serotonin neurotransmitter systems, there are also abnormalities in multiple neurotransmitter systems including choline, norepinephrine, histamine, and gamma-aminobutyric acid in SCZ, and currently there are antipsychotics targeting the above neurotransmitter systems in development or clinical trials. Therefore, we urgently need to develop a feasible and effective method to achieve simultaneous detection of multiple neurotransmitter systems, so as to analyze the heterogeneity related to the neurotransmitter system in SCZ and support the development of future individualized and precise drug treatment plans.

[0004] Previous studies have demonstrated the association between neurotransmitter systems and brain structures: Zhen et al. found a correlation between the co-variation network of cortical thickness (CTh) and the co-variation network of neurotransmitter systems; Liao et al. found that the decrease in CTh during neurodevelopment reflects the overall organizational characteristics of the neurotransmitter system in the brain; Liang et al. found that the decrease in CTh in SCZ is related to the increase in glutamate concentration in the dorsal anterior cingulate region. These research results suggest that in SCZ, brain structure abnormalities, especially CTh abnormalities, can indirectly reflect neurotransmitter system abnormalities. On this basis, we propose the following methodological pipeline for analyzing the heterogeneity of neurotransmitter-related SCZ, that is, first achieving individualized measurement of CTh abnormalities in SCZ patients, then associating the neurotransmitter system with individual CTh abnormalities, thereby indirectly measuring the abnormalities of the neurotransmitter system in individual SCZ patients. On this basis, biological subtypes are separated, so as to explore the genetic and neuropathological mechanisms of different subtypes of SCZ from multiple neurotransmitter systems, providing a scientific basis and reference for the selection of individualized and precise antipsychotic drugs in clinical practice.

[0005] In recent years, the development of imaging transcriptomics technology has brought more possibilities to brain imaging research. By associating disease imaging findings with publicly available multimodal data, such as gene expression in the brain, PET neurotransmitter receptor / transporter maps, or brain microstructure maps, we can infer the genetic, molecular, and cellular tissue characteristics corresponding to brain structure and function abnormalities in diseases. Using this method, Park et al. found that the brain structure abnormalities shared by multiple mental diseases are related to the dopamine and serotonin neurotransmitter systems. The applicant also found an association between the dominant brain function impairment pattern in SCZ and the brain distribution of D2 and 5-HT2a receptors in previous studies. Combining the above methods with brain structure development norm construction technology, we can associate brain structure abnormalities with multiple neurotransmitter systems at the individual patient level. Based on individual brain structure deviations marked by neurotransmitters, we can identify neurotransmitter-related SCZ biological subtypes and analyze the heterogeneity of SCZ from the core treatment target of the neurotransmitter system. Biase et al. successfully separated SCZ biological subtypes from the perspective of CTh-related brain cell types based on a similar research idea.

[0006] In summary, there are problems in current SCZ research, such as large interference from disease heterogeneity, difficulty in detecting multiple neurotransmitter systems in individual patients, and overlapping neuropathological and genetic mechanisms corresponding to clinical manifestations. Existing research techniques do not solve the problem of simultaneously measuring multiple neurotransmitters, cannot obtain the abnormal spectrum of multiple neurotransmitter systems in a single individual, and it is difficult to use individualized neurotransmitter abnormality information to classify subtypes of schizophrenia.

[0007] Previous studies have mainly analyzed the disease heterogeneity of SCZ from two perspectives: symptomatic subtypes and biological subtypes. Zhang et al. and Nenadic et al. both classified SCZ into three symptomatic subtypes, namely positive (paranoid) symptom subtype, negative symptom subtype, and disorganized symptom subtype, according to clinical symptom manifestations, and found differences in brain structural abnormalities among different subtypes, but their results were inconsistent. Xiao et al. classified SCZ into three biological subtypes with different brain structural abnormalities according to brain surface area, volume, and CTh. Pan et al. also classified SCZ into three biological subtypes according to CTh, but no differences in clinical symptom manifestations were found among different biological subtypes. Chai et al. systematically compared the classification methods of symptomatic and biological subtypes of SCZ and found that the subtypes obtained by the two methods did not match, as Figure 1 shown. The above studies indicate that the same clinical symptom manifestations in SCZ may have very different biological bases. Starting from biological subtypes, the genetic and neuropathological bases of SCZ heterogeneity can be more accurately analyzed.

[0008] As Figure 2 shown: In previous studies, the identification of biological subtypes often directly used biological indicators of SCZ patients (such as CTh) for classification. Varol et al. were the first to point out that this method may introduce factors other than the disease effect, such as individual differences, age, and gender effects that also exist in the normal population. Based on this, they proposed the "Heterogeneity through Discriminative Analysis (Hydra)" method. This method introduces normal control information in the process of subtype identification and identifies subtypes based on the pure disease effect. Chand et al. used the Hydra method to classify SCZ into two biological subtypes, and this subtype classification method has been replicated in multiple subsequent independent studies, demonstrating the superiority of this research idea. However, it should be noted that the separation process of the disease effect by Hydra occurs in the high-dimensional space of the machine learning model. At the individual patient level, the overall disease effect can only be characterized by the abstract "signature presence", which cannot reflect the impact of the disease effect on specific brain regions and is not conducive to further exploring the neuropathological mechanisms of specific subtypes. The introduction of the norm method well solves the above problems. Similar to the common height or weight development norms, by using neuroimaging big data to construct brain structure development norms, we can obtain the normal ranges of brain structure indicators corresponding to specific ages, genders, and other variables. Then, by comparing the brain structure indicators of SCZ patients with their corresponding norms, we can individually measure the deviation between the brain structure of each brain region and the normal value, obtain the brain structure deviation map of a single patient, separate the disease effect at the individual level, and use it for subsequent subtype identification.

[0009] Disadvantages of the prior art:

[0010] (1) The molecular imaging detection method of neurotransmitter systems based on Positron Emission Tomography (PET) can only detect one neurotransmitter receptor / transporter in one imaging. However, existing evidence shows that in addition to the dopamine and serotonin neurotransmitter systems, there are abnormalities in multiple neurotransmitter systems in SCZ, including choline, norepinephrine, histamine, and gamma-aminobutyric acid, etc. Moreover, each neurotransmitter system contains multiple neurotransmitter transporter / receptor subtypes. In addition, PET operation also has problems such as invasiveness, risk of drug allergy, and high economic burden. These defects make it extremely difficult and have very low feasibility to measure abnormalities in multiple neurotransmitter systems in the same individual using PET technology at the same time.

[0011] (2) The method of heterogeneity analysis based on discriminant analysis (Heterogeneity through Discriminative Analysis, Hydra) separates the disease effect in the high-dimensional space of the machine learning model. At the individual patient level, the overall disease effect can only be characterized by the abstract "signature presence", which cannot reflect the impact of the disease effect on specific brain regions, is not conducive to further exploring the neuropathological mechanisms of specific subtypes, and localizing abnormal brain regions at the individual level.

[0012] (3) Traditional clustering methods such as K-means clustering or hierarchical clustering have problems such as high randomness, poor stability, and poor repeatability between different data sets. The models obtained using the above clustering methods have poor transferability, and using such models is not conducive to identifying stable and robust neurotransmitter-related SCZ subtypes on different sites / different magnetic resonance scanners. Summary of the Invention

[0013] Based on the above problems, the object of the present invention is to provide a method for classifying neurotransmitter subtypes of schizophrenia, which indirectly reflects the possible abnormalities of the neurotransmitter system in individual SCZ patients by associating publicly available PET neurotransmitter receptor or transporter maps with individual brain structure deviations, and taking this as a starting point, analyzes the heterogeneity of SCZ disease from the perspective of multiple neurotransmitter systems, and identifies neurotransmitter-related SCZ biological subtypes.

[0014] The technical solution adopted by the present invention to achieve its invention object is a method for classifying neurotransmitter subtypes of schizophrenia, including the following steps:

[0015] S1. Using the brain structural MRI data of multiple healthy controls, train the cortical thickness (CTh) brain structural development norm using the Hierarchical Bayesian Regression (HBR) method. This norm is an HBR model related to the predicted values of the cortical thickness CTh in each brain region.

[0016] In specific implementation, in addition to the HBR method, techniques such as the general linear model, Gaussian process regression, and additive model can also be used to construct the CTh brain structural development norm.

[0017] S2. Collect the magnetic resonance data of multiple schizophrenia patients, and then extract the measured values of the cortical thickness CTh in each brain region of each patient.

[0018] S3. Using the HBR model trained in step S1, calculate the CTh deviation Z value of each brain region of each patient through the following formula:

[0019]

[0020] In the formula: y is the measured value of CTh of the patient, is the CTh value predicted by the HBR model, σ is the prediction variance, and σ n is the internal variance of the HBR model.

[0021] S4. Collect the PET molecular imaging maps of a certain neurotransmitter receptor or transporter A (A is a positive integer), calculate the mean value of each brain region, and then use Spearman correlation to calculate the correlation coefficient ρ between each PET molecular imaging map and the individual CTh deviation of the patient. Then, form the neurotransmitter spectrum related to CTh abnormality by combining the Spearman correlation coefficients ρ of all A PET molecular imaging maps and the individual CTh deviation of the patient.

[0022] In specific implementation, techniques such as Pearson correlation and cosine correlation can also be used to replace Spearman correlation to associate the individual brain structure deviation map with the neurotransmitter map and construct the neurotransmitter spectrum related to individual CTh abnormality.

[0023] S5. Reduce the dimension of the neurotransmitter spectrum related to CTh abnormality obtained in step S4, reducing the dimension from A to 3.

[0024] S6. Randomly sample the healthy control brain structural MRI data set B times (B is a positive integer) at a set sampling rate, perform K-means clustering on each randomly sampled data set, the range of the number of clusters is from 2 to C (C is a positive integer greater than 2), perform consistency clustering for each number of clusters, then evaluate and determine the optimal number of clusters, that is, the number of subtypes, and then select the clustering model corresponding to the number of subtypes as the final schizophrenia neurotransmitter subtype classification model.

[0025] In specific implementation, other clustering methods besides consistency clustering can also be used to achieve subtype identification and model construction;

[0026] S7. Collect the magnetic resonance data of schizophrenia patients to be typed, extract the measured CTh values of each brain region of the patient, then construct the neurotransmitter spectrum related to CTh abnormality of the patient by the method in step S4, and input it into the final schizophrenia neurotransmitter subtype classification model obtained in step S6 to obtain the neurotransmitter-related subtype of the patient.

[0027] Furthermore, the HBR model in step S1 is as follows:

[0028]

[0029] In the formula: Y is the predicted CTh value of each brain region, f represents the HBR model function, X is the input data, including age, gender, type of MRI scanner, and average cortical thickness of the cerebral hemisphere, θ is the model parameter obtained after training the model, is the model residual.

[0030] Furthermore, in step S1, before formal training, hierarchical Bayesian regression method is first used for pre-training, and then the HBR model is trained by five-fold cross-validation method. After cross-validation determines the performance of the HBR model, formal training is carried out.

[0031] Furthermore, the specific Spearman correlation calculation in step S4 is as follows:

[0032] First, the two types of data, namely the CTh deviation value Z of each brain region of each patient and the PET molecular imaging map value of the neurotransmitter receptor or transporter, are respectively converted into rank values, and then the Spearman correlation coefficient ρ is calculated using the following formula:

[0033]

[0034] In the formula: n is the number of data samples in each category, and d is the interpolation of the ranks of a pair of data;

[0035] The specific method for converting the two types of data into rank values is to assign values in ascending order as 1, 2, 3,....

[0036] Furthermore, the specific method for dimensionality reduction in step S5 is: use the uniform manifold approximation and projection method, namely the UMAP method, for dimensionality reduction.

[0037] Furthermore, the specific method for evaluating and determining the optimal number of clusters, that is, the number of subtypes, in step S6 is: evaluate the increase in the fuzzy clustering ratio score, average silhouette index, and area under the cumulative distribution function of the consensus matrix, and take the number of clusters pointed to by the three indicators as the most to be determined as the optimal number of clusters, that is, the number of subtypes.

[0038] Further, the number of the brain regions is 68, and the DK (Desikan-Killiany) brain region segmentation template of the prior art is adopted for the division of the brain regions.

[0039] Further, A is equal to 21, and the PET molecular imaging atlas of the A neurotransmitter receptors or transporters includes the PET molecular imaging atlases of 21 neurotransmitter receptors or transporters in a total of 9 neurotransmitter systems, specifically including:

[0040] Dopamine system: Receptor D1, receptor D2, transporter DAT, and neuron dopamine uptake ability marker FDOPA;

[0041] Serotonin system: Receptor 5-HT1a, receptor 5-HT1b, receptor 5-HT2a, receptor 5-HT4, receptor 5-HT6, and transporter 5-HTT;

[0042] Cannabinoid system: Receptor CB1;

[0043] Opioid system: Receptor MOR, receptor KOR;

[0044] Glutamate system: Receptor mGluR5, receptor NMDA;

[0045] Gamma-aminobutyric acid system: Receptor GABAa;

[0046] Histamine system: Receptor H3;

[0047] Adrenergic system: Transporter NET;

[0048] Cholinergic system: Receptor M1, receptor α4β2, and transporter VAChT.

[0049] At this time, in step S5, the dimension of the neurotransmitter spectrum related to CTh abnormality is reduced from 21 to 3.

[0050] Further, in step S6, B is equal to 100, that is, the previously collected healthy control brain structure MRI data set is randomly sampled 100 times at the set sampling rate.

[0051] Further, in step S6, C is equal to 8, that is, the set clustering number range is from 2 to 8.

[0052] The beneficial effects of the present invention are:

[0053] (1) Multiple neurotransmitter labeling was performed on the brain structural abnormalities of SCZ patients to indirectly reflect the abnormalities of the neurotransmitter system, and innovatively achieved the simultaneous measurement of multiple neurotransmitter systems at the individual level. The abnormality of the neurotransmitter system is the core neuropathological mechanism and treatment target of SCZ. However, due to the high cost, invasive operation and risk of drug allergy of PET testing, it is difficult to use PET technology to simultaneously detect multiple neurotransmitter systems in a single project. By correlating the publicly available PET neurotransmitter receptor / transporter atlas with the individual brain structural deviation, this invention indirectly reflects the possible abnormalities of the neurotransmitter system in SCZ patients, and takes this as a starting point to analyze the disease heterogeneity of SCZ from the perspective of multiple neurotransmitter systems and identify the SCZ biological subtypes related to neurotransmitters.

[0054] (2) By using the currently publicly available human brain imaging data and the previously collected data, a relatively large sample size was achieved. The nonlinear modeling of the brain development process can be carried out while dealing with the possible multi-center bias in data processing by using the HBR method. With the above data and modeling scheme, a CTh brain structure development norm with good performance can be obtained.

[0055] (3) Using the consistency clustering method, the clustering scheme with the highest consistency was obtained through repeated sampling and clustering, which can effectively improve the stability and repeatability of clustering analysis.

[0056] (4) Through the consistency clustering of the neurotransmitter spectra related to the individual-level structural abnormalities of the subjects and verification in multiple independent samples, stable and reliable SCZ-related neurotransmitter-related subtypes were obtained, realizing a subtype classification closer to the biological basis of SCZ. Description of the Drawings

[0057] Figure 1 It is a schematic diagram of the mismatch between the symptomatic subtypes and biological subtypes of SCZ;

[0058] Figure 2 It is a schematic diagram of the two traditional subtype classification methods and the subtype classification method of this invention;

[0059] Figure 3 It is the PET molecular imaging atlas of neurotransmitter receptors or transporters collected in the embodiments of this invention;

[0060] Figure 4 It is the schematic diagram of the flow of the embodiments of this invention. Detailed Description of the Invention

[0061] In order to more clearly understand the above objects, features and advantages of the present invention, the present invention will be further described in detail below with reference to the drawings and specific embodiments.

[0062] Figure 4A specific implementation method for typing neurotransmitter subtypes of schizophrenia in the present invention is shown, and its steps include:

[0063] S1. Construct a CTh brain structure development norm, specifically: Use at least 2000 publicly available and previously collected healthy control brain structure MRI data. Use the "estimate" command in the mature toolkit PCNtoolkit for constructing the brain development norm, select the Hierarchical Bayesian Regression (HBR) method, and adopt the five-fold cross-validation method to train the HBR model. After cross-validation determines the performance of the HBR model, use at least 2000 publicly available and previously collected healthy control brain structure MRI data again, and use the hierarchical Bayesian regression method to train the HBR model to obtain the model parameter set θ; The HBR model formula is as follows:

[0064]

[0065] In the formula: Y is the CTh prediction value of each brain region, f represents the HBR model function, X is the input data, including age, gender, MRI scanner type, and the average cortical thickness of the cerebral hemisphere, θ is the model parameter obtained after training the model, is the model residual; CTh represents the cortical thickness;

[0066] The number of the brain regions is 68, and the division of the brain regions adopts the existing DK (Desikan-Killiany) brain region segmentation template;

[0067] S2. Use a 3.0 T magnetic resonance scanner to collect high-resolution T1-weighted structural magnetic resonance data of at least 100 schizophrenia patients. The spatial resolution needs to be less than 1mm×1mm×1mm per voxel. After using the existing dcm2niix toolkit to convert the original DICOM format image data exported from the magnetic resonance scanner into NIFTI format data for analysis, then use the "recon_all" command in the existing Freesurfer software to process it, and use the "aparc2table" command to extract the cortical thickness (Cortical thickness, CTh) data of 68 brain regions of each patient, that is, the CTh measured value;

[0068] S3. Use the HBR model that has been trained in step S1 to calculate the CTh deviation, that is, the Z value, of each brain region of each patient through the following formula:

[0069]

[0070] In the formula: y is the CTh measured value of the patient, The CTh value predicted by the HBR model, σ is the predicted variance, σ n is the internal variance of the HBR model, reflecting the internal variation of CTh values in the healthy control group;

[0071] S4. Collect the PET molecular imaging maps of 21 neurotransmitter receptors or transporters in 9 existing neurotransmitter systems, such as Figure 3 shown, specifically including:

[0072] Dopamine system: Receptor D1, Receptor D2, Transporter DAT, Neuron dopamine uptake ability marker FDOPA;

[0073] Serotonin system: Receptor 5-HT1a, Receptor 5-HT1b, Receptor 5-HT2a, Receptor 5-HT4, Receptor 5-HT6, Transporter 5-HTT;

[0074] Cannabinoid system: Receptor CB1;

[0075] Opioid system: Receptor MOR, Receptor KOR;

[0076] Glutamate system: Receptor mGluR5, Receptor NMDA;

[0077] γ-aminobutyric acid system: Receptor GABAa;

[0078] Histamine system: Receptor H3;

[0079] Adrenergic system: Transporter NET;

[0080] Cholinergic system: Receptor M1, Receptor α4β2, Transporter VAChT;

[0081] After each PET molecular imaging map is registered to the standard space, the mean value of each brain region is calculated, and then the Spearman correlation is used to calculate the correlation coefficient ρ between each PET molecular imaging map and the deviation of the patient's individualized CTh. In the calculation of the Spearman correlation coefficient, the CTh deviation values of the 68 brain regions of each patient and the PET molecular imaging map values of the neurotransmitter receptor or transporter are first converted into rank values, and then the following formula is used to calculate the Spearman correlation coefficient:

[0082]

[0083] In the formula: ρ is the Spearman correlation coefficient, n is the number of data samples in each category, and d is the interpolation of the ranks of a pair of data;

[0084] The specific method for converting the two types of data into rank values is to assign values in ascending order as 1, 2, 3,...;

[0085] Next, the Spearman correlation coefficient ρ between the PET molecular imaging maps of all 21 neurotransmitter receptors or transporters and the individual patient's CTh deviation is used to form a neurotransmitter spectrum related to CTh abnormality, and this is used to label the neurotransmitters in the individual patient's CTh deviation map. This vector is used for subsequent clustering analysis to identify neurotransmitter-related SCZ biological subtypes;

[0086] S5. Use the Uniform Manifold Approximation and Projection (UMAP) method for dimensionality reduction, reducing the dimension of the neurotransmitter spectrum related to CTh abnormality obtained in step S4 from 21 to 3. Dimensionality-reduced data can improve clustering performance, and the UMAP method is used for dimensionality reduction because UMAP performs better than commonly used component analysis or independent component analysis methods in preserving the local and global structure of the data. The parameters of UMAP are set according to its guidelines, with the number of nearest neighbors set to 30 and the minimum distance set to 0;

[0087] S6. After dimensionality reduction, perform consensus clustering using the "run_all_consensus_partition_method" command in the Cola package in R. Randomly sample the previously collected healthy control brain structure MRI dataset 100 times at an 80% sampling rate, and perform K-means clustering on each randomly sampled dataset. The number of clusters ranges from 2 to 8, and for each number of clusters, perform the consensus clustering method, that is, create multiple clustering schemes and then represent the frequency of each pair of data points being assigned to the same group in the 100 clustering schemes by constructing a consensus matrix. Determine the clustering scheme for each number of clusters by selecting the maximum frequency;

[0088] After completing the consensus clustering for the number of clusters from 2 to 8 respectively, evaluate three metrics, where:

[0089] The higher the score of the Proportion of Ambiguous Clustering (PAC), the better;

[0090] The higher the score of the Mean Silhouette Index, the better;

[0091] The increase in the Area Under the Cumulative Distribution Function of the Consensus Matrix, and determine the last number of clusters before a significant drop in the increased value;

[0092] The clustering quantity with the most pointing of the three indicators is determined as the optimal clustering quantity, that is, the subtype quantity, and then the clustering model corresponding to the subtype quantity is selected as the final neurotransmitter subtype classification model for schizophrenia;

[0093] S7. Collect the magnetic resonance data of the schizophrenia patients to be classified, extract the cortical thickness data of each brain region of the patient, that is, the measured CTh value, and then adopt the method in step S4 to construct the neurotransmitter spectrum related to the CTh abnormality of the patient, and input it into the final neurotransmitter subtype classification model of schizophrenia obtained in step S6 to obtain the neurotransmitter-related subtype type of the schizophrenia patient.

[0094] The above embodiments of the present invention are only examples for explaining the present invention, rather than limiting the implementation manners of the present invention. For those of ordinary skill in the art, other different forms of changes and modifications can be made on the basis of the above description. It is impossible to list all the implementation manners here. Any obvious changes or modifications derived from the technical solutions of the present invention still fall within the protection scope of the present invention.

Claims

1. A method for classifying neurotransmitter subtypes of schizophrenia, characterized in that: The steps include: S1. Using brain structure MRI data of multiple healthy controls, a hierarchical Bayesian regression method was used to train the CTh brain structure development norm, i.e., the HBR model related to the predicted value of CTh of the cortical thickness of each brain region; the HBR model is as follows: ； Where: Y is the predicted CTh value of each brain region, f represents the HBR model function, X is the input data, including age, gender, MRI scanner type and average cortical thickness of the cerebral hemisphere, θ is the model parameter obtained after training the model, is the model residual; S2, collecting magnetic resonance data of multiple schizophrenia patients, and then extracting the measured value of cortical thickness CTh of each brain region of each patient; S3. Using the HBR model trained in step S1, calculate the CTh deviation Z value of each brain region of each patient by the following formula: ； Where: y is the actual measured value of CTh of the patient, is the CTh value predicted by the HBR model, σ is the prediction variance, σ n is the internal variance of the HBR model; S4. Collect A kinds of PET molecular imaging spectra of neurotransmitter receptors or transporters and calculate the mean of each brain region, where A is a positive integer, and then use Spearman correlation to calculate the correlation coefficient ρ between each PET molecular imaging spectra and the individualized CTh deviation of the patient, and then the Spearman correlation coefficient ρ of all A kinds of PET molecular imaging spectra and the individualized CTh deviation of the patient is combined into a CTh abnormality-related neurotransmitter spectrum; S5, reducing the dimension of the CTh abnormality-related neurotransmitter spectrum obtained in step S4 from A to 3; S6. Randomly sample the healthy control brain structure MRI data set B times at a set sampling rate, where B is a positive integer, and perform K-means clustering on each randomly selected data set, where the number of clusters ranges from 2 to C, where C is a positive integer greater than 2, and perform consistency clustering on each number of clusters, and then evaluate and determine the optimal number of clusters, i.e., the number of subtypes, and then select the clustering model corresponding to the number of subtypes as the final schizophrenia neurotransmitter subtype classification model; S7. Collect magnetic resonance data of the schizophrenia patient to be classified, extract the CTh measured value of each brain region of the patient, and then use the method of step S4 to construct the patient's CTh abnormality-related neurotransmitter spectrum, and then input the final schizophrenia neurotransmitter subtype classification model obtained in step S6 to obtain the patient's neurotransmitter-related subtypes.

2. The method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: In step S1, before formal training, the hierarchical Bayesian regression method is used for pre-training, and then the HBR model is trained using the five-fold cross-validation method. After the cross-validation determines the performance of the HBR model, formal training is performed.

3. A method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: The Spearman correlation calculation in step S4 is specifically as follows: First, the CTh deviation value Z of each brain region of each patient and the PET molecular imaging atlas value of the neurotransmitter receptor or transporter are converted into rank values ​​respectively, and then the Spearman correlation coefficient ρ is calculated using the following formula: Where: n is the number of samples of each type of data, d is the interpolation of a pair of data ranks; The specific method of converting the two types of data into rank values ​​is to assign values ​​1, 2, 3, ... in ascending order.

4. The method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: The specific method of dimensionality reduction in step S5 is: using the uniform manifold approximation and projection method, namely the UMAP method, to perform dimensionality reduction.

5. The method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: The specific method for evaluating and determining the optimal number of clusters, i.e., the number of subtypes, described in step S6 is: evaluating the increase in the fuzzy clustering ratio score, the average silhouette index, and the area under the cumulative distribution function of the consensus matrix, and taking the number of clusters that the three indicators point to the most as the optimal number of clusters, i.e., the number of subtypes.

6. The method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: The number of brain regions is 68, and the brain regions are divided using the DK brain region segmentation template.

7. The method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: A is equal to 21, and the PET molecular imaging atlas of the A neurotransmitter receptors or transporters includes PET molecular imaging atlases of 21 neurotransmitter receptors or transporters in 9 neurotransmitter systems, specifically including: Dopamine system: receptor D1, receptor D2, transporter DAT, neuronal dopamine uptake capacity marker FDOPA; Serotonin system: receptor 5-HT1a, receptor 5-HT1b, receptor 5-HT2a, receptor 5-HT4, receptor 5-HT6, transporter 5-HTT; Cannabinoid system: receptor CB1; Opioid system: receptor MOR, receptor KOR; Glutamate system: receptor mGluR5, receptor NMDA; Gamma-aminobutyric acid system: receptor GABAa; Histamine system: receptor H3; Adrenaline system: transporter NET; Cholinergic system: receptor M1, receptor α4β2, transporter VAChT.

8. The method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: In step S6, B is equal to 100.

9. The method for classifying neurotransmitter subtypes of schizophrenia according to claim 1, characterized in that: In step S6, C is equal to 8.

Images