Construction method, system and equipment of depression prognosis prediction model, medium and application
Through the combination of PET scanning and T1-weighted MRI imaging, combined with support vector machine algorithm, a depression prognosis prediction model based on mGluR5 receptor availability was constructed, solving the problem of measurement instability and lack of biomarkers in the prior art, and realizing the High sensitivity and specific prognosis prediction, supporting individualized treatment.
Patent Information
- Application Number
- CN202411250234.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-06
- Publication Date
- 2024-12-27
AI Technical Summary
The existing prediction models for prognosis of depression have measurement instability and lack of effective biomarkers, resulting in poor prediction results, especially in the low reliability of repeated measurements based on BOLD signals.
PET scanning and T1-weighted MRI imaging were combined to detect the availability of mGluR5 receptors by injecting tracer labeled with radioisotopes [18F]FPEB, and a support vector machine algorithm was used to construct a prediction model for prognosis of depression, optimized feature sets and model performance. Select the best model as the prognostic prediction model.
Improve measurement stability and reliability of repeated measurements, provide a more accurate biological basis, enhance the sensitivity and specificity of prognostic prediction, and support individualized medical strategies. The model is shown as an accuracy rate of 0.88 and a recall rate of 0.93. Accuracy is 0.87 and F1 score is 0.90.
Smart Images
Figure CN119207790A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of depression prognosis prediction, and in particular relates to a method, system, device, medium for constructing a depression prognosis prediction model, and the application of metabotropic glutamate receptor 5 (mGluR5) as a biomarker in constructing a depression prognosis prediction model. Background Art
[0002] As a common mental health problem in the world, depression has long been the focus of clinical treatment and scientific research. At present, the imaging-based depression prognosis prediction models on the market can be roughly divided into two categories: one is based on structural imaging (such as T1-weighted imaging, diffusion tensor imaging [DTI], etc.) for prognosis prediction; the other is based on functional imaging (such as blood oxygen level-dependent signal imaging [BOLD]) for prognosis prediction. However, there are limitations in using prognosis prediction models based on structural imaging, because not all patients with depression show obvious brain structural abnormalities, which makes this type of model unsuitable for some patients. The functional imaging model that relies on the BOLD signal is limited by the inherent instability of the BOLD signal itself and the low reliability of repeated measurements, which makes the measurement results in practical applications unstable, which also affects its effectiveness and consistency in predicting the prognosis of depression.
[0003] At present, the role of the glutamate system in the diagnosis and prediction of MDD has received increasing attention. Metabotropic glutamate receptor 5 (mGluR5) is a Gq / 11G protein-coupled receptor that is mainly expressed at the postsynaptic sites of neurons. Recent studies have shown that it is closely related to the pathogenesis of depression. Studies have found that mGluR5 may play a role in the development of depression by regulating neurotransmitter balance and synaptic plasticity. Animal experiments have also shown that reducing the expression of mGluR5 affects the expression of genes related to stress coping and emotional stability, leading to depressive-like behavior under long-term stress. Conversely, enhancing the expression of mGluR5 can improve the resistance of mice to stress. In addition, previous studies have repeatedly pointed out that the reduced expression of mGluR5 in patients with depression is related to the pathophysiology of the disease. GWAS studies and autopsy studies have shown that the expression level of Grm5 (the encoding gene of mGLUR5) / mGLUR5 in patients with depression is low. Molecular imaging studies have also shown that the availability of mGluR5 in patients with MDD is significantly reduced. In terms of antidepressant drug treatment, animal experiments have shown that mice with low mGluR5 expression are resistant to treatment. However, it is unclear whether testing mGLUR5 receptor availability can predict depression prognosis in a real clinical setting. Summary of the invention
[0004] The technical problem to be solved by the present invention is to overcome the deficiencies and defects mentioned in the above background technology, provide a method for constructing a depression prognosis prediction model, a system for constructing a depression prognosis prediction model, corresponding electronic devices and non-transitory computer-readable storage media, and also provide the application of mGluR5 as a biomarker in constructing a depression prognosis prediction model.
[0005] In order to solve the above technical problems, the technical solution proposed by the present invention is:
[0006] In a first aspect, the present invention provides a method for constructing a depression prognosis prediction model, comprising the following steps:
[0007] (1) Obtaining PET images and T1-weighted MRI images: PCT images of mGLUR5 receptor availability in different patients were obtained through PET scanning, and then T1-weighted MRI images were obtained;
[0008] (2) Image preprocessing: The mGluR5 receptor availability PET images of different patients obtained in step (1) were registered with the corresponding T1-weighted MRI images, the T1-weighted MRI images were segmented using statistical parameter mapping software, and the non-displacement binding potential of mGluR5 availability was calculated using a simplified reference tissue model with the cerebellum as the reference area;
[0009] (3) Model construction: After the image preprocessing in step (2) is completed, the mGLUR5 receptor availability of each brain region is extracted through the AAL116 brain atlas, and the model is constructed using RFE technology and the features that contribute the least to the prediction target are gradually removed to optimize the feature set. This process is repeated until the preset number of features or model performance threshold is reached; a preliminary model is constructed using the support vector machine algorithm, the model is initialized, the model performance is analyzed and optimized, and the model with the best overall effect is selected as the depression prognosis prediction model.
[0010] In the above-mentioned construction method, preferably, the process of step (1) specifically includes the following steps: intravenous injection of the radiotracer [18F]FPEB into the patient group, followed by collection of balanced emission data on a PET / CT scanner, using the obtained data using a TOF (Time-of-Flight) algorithm to locate the position of the radiotracer [18F]FPEB in the body, and reconstructing a mGLUR5 receptor availability PET image, and then using a 32-channel head coil on a 3T magnetic resonance scanner to acquire a T1-weighted MRI image to assess potential structural abnormalities and assist in registering the PET data to the standard space.
[0011] The present invention is the first to test whether mGLUR5 receptor availability can predict the prognosis of depression in a real clinical environment. However, there are two major problems in predicting the prognosis through mGLUR5 receptor availability: 1) the stability problem of mGluR5 receptor measurement; 2) the problem of constructing a normalized model of neuroimaging based on receptor availability. In order to overcome the problem of constructing a normalized model of mGluR5 receptor neuroimaging and comprehensively evaluate the expression of mGluR5 in different patient groups, the present invention uses PET scanning and T1-weighted MRI imaging. In order to overcome the stability problem of mGluR5 receptor measurement, the present invention detects the biochemical activity inside the body by injecting a tracer [18F]FPEB labeled with a radioactive isotope. When the positrons released by the tracer collide with electrons, gamma rays are generated, and the PET scanner constructs images by capturing these rays. At the same time, T1-weighted MRI is a technology that obtains images of in vivo structures by adjusting magnetic fields and radio frequency pulses. Its characteristic is that the image is sensitive to the T1 relaxation time by adjusting the scanning parameters, thereby obtaining high-contrast anatomical details. The purpose of acquiring T1-weighted MRI images is to accurately align them with PET images to ensure the precise matching of anatomical structures and to more accurately locate the distribution of mGluR5.
[0012] Preferably, in step (2), the registration process is as follows: first, the images of the two different modalities are denoised and image deviations are corrected, and the optimal registration parameters are determined by maximizing the common information between the mGLUR5 receptor availability PET image and the MRI image, and then manual quality control is performed to adjust and optimize the registration results to ensure that the anatomical structures between different images can be accurately aligned.
[0013] Preferably, in step (2), the process of segmenting the T1-weighted MRI image using statistical parameter mapping software includes the following steps: segmenting the T1-weighted MRI image using statistical parameter mapping software to obtain gray matter and white matter probability masks representing different brain tissues; then, spatially normalizing the T1 image using the ANTS software package to achieve accurate alignment by minimizing the difference between the individual brain anatomical structure and the standard brain template; finally, fine-tuning the shape of individual brain regions to match the standard template to the greatest extent; after the T1 image normalization is completed, applying the obtained transformation parameters to the corresponding PET image; after the alignment is completed, verifying the accuracy of the alignment through visual inspection and automated quality assessment steps based on MRICRON and MATLAB.
[0014] Preferably, in step (3), after extracting the mGLUR5 receptor availability of each brain region through the AAL116 brain atlas, a univariate analysis is performed on each feature using the t-test statistical test to screen out features that are significantly correlated with the prognosis of depression. Based on the preliminary t-test screening, a recursive feature elimination technique is used to select features more finely.
[0015] Preferably, in step (3), a preliminary model is constructed using a support vector machine algorithm, and the model is initialized using Python's scikit-learn machine learning library; before training the support vector machine algorithm model, a standardized method is used for feature scaling, that is, all features are scaled to a range of a mean of 0 and a variance of 1, to ensure that the model has a unified evaluation standard in all dimensions.
[0016] Preferably, in step (3), the process of analyzing and optimizing the model performance is as follows: a leave-one-out cross-validation method is used to evaluate the model performance under different parameter combinations; in each leave-one-out cross-validation iteration, the accuracy, recall, precision, and F1 score of the model are recorded and calculated; the receiver operating characteristic curve and the area under the curve of the model are calculated to evaluate the overall performance of the model; and then the performance indicators under different parameter settings are analyzed, and the model with the best comprehensive effect is selected as the depression prognosis prediction model.
[0017] In a second aspect, the present invention provides a system for constructing a depression prognosis prediction model, comprising:
[0018] A data acquisition module, used to acquire mGLUR5 receptor availability PET images of different patients through PET scanning, and then acquire T1-weighted MRI images;
[0019] An image preprocessing module to register the mGLUR5 receptor availability PET images of different patients with the corresponding T1-weighted MRI images, segment the T1-weighted MRI images using statistical parameter mapping software, and calculate the non-displacement binding potential of mGluR5 availability using a simplified reference tissue model with the cerebellum as the reference region;
[0020] A model building module is used to extract the mGLUR5 receptor availability of each brain region through the AAL116 brain atlas, use RFE technology to build a model and gradually remove the features that contribute the least to the prediction target to optimize the feature set. This process is repeated until the preset number of features or model performance threshold is reached; a preliminary model is built using a support vector machine algorithm, the model is initialized, the model performance is analyzed and optimized, and the model with the best comprehensive effect is selected as the depression prognosis prediction model.
[0021] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the construction method when executing the program.
[0022] In a fourth aspect, the present invention provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the construction method.
[0023] In a fifth aspect, the present invention provides an application of mGluR5 as a biomarker in constructing a depression prognosis prediction model. In view of the core role of mGluR5 in the pathogenesis of depression, its receptor status can provide direct information about the disease state and treatment response, and can provide higher sensitivity and specificity in prognosis prediction, which has not been fully utilized in the existing prognosis prediction models on the market.
[0024] Preferably, the above application uses mGluR5 as a biomarker, obtains mGluR5 receptor availability data through PET scanning, and constructs a depression prognosis prediction model based on the mGluR5 receptor availability data.
[0025] Compared with the prior art, the present invention has the following beneficial effects:
[0026] 1. The method for constructing a depression prognosis prediction model of the present invention is based on advanced molecular imaging technology and focuses on the precise measurement of changes in the availability of mGluR5). The constructed depression prognosis prediction model has the following significant advantages over the existing imaging-based depression prognosis prediction models on the market: 1) Measurement stability and reliability of repeated measurements: This model uses PET scanning to ensure the high stability and retest reliability of the measurement data, effectively avoiding the variability and uncertainty commonly seen in traditional functional imaging technologies based on BOLD signals; 2) A highly targeted biomarker: mGluR5 not only plays a core role in the pathophysiology of depression, but also changes in its availability directly reflect the disease state and treatment response. This highly targeted biomarker provides a more accurate biological basis for prognosis prediction; 3) High sensitivity and specificity: Due to its direct link to the key neurobiological mechanisms of depression, this model shows high sensitivity and specificity in identifying changes in disease status and predicting treatment responses; 4) Personalized medicine: By accurately evaluating the availability of mGluR5 receptors, this technology makes it possible to implement personalized treatment strategies, enabling doctors to adjust treatment plans and estimate prognostic effects based on each patient's specific biomarkers, which has significant innovation and application value.
[0027] 2. The present invention utilizes the changes in the receptor availability of mGluR5 to construct a prognostic model for patients with depression. By adopting PET scanning technology, the availability of mGluR5 is accurately measured. Compared with the existing functional imaging-based methods, this technology has higher measurement stability and reliability. It can also provide more personalized treatment plans for patients with depression based on the evaluation of mGluR5 availability, which is a major innovation in the existing prognostic prediction models.
[0028] 3. The depression prognosis prediction model constructed by the present invention has a model accuracy of 0.88, a recall rate of 0.93, a precision of 0.87, and an F1 score of 0.90, with good performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0030] Figure 1 This is a flow chart of the depression prognosis prediction model based on mGluR5;
[0031] Figure 2 It is the accuracy, recall, precision and F1 score of the depression prognosis prediction model based on mGluR5. DETAILED DESCRIPTION
[0032] In order to facilitate the understanding of the present invention, the present invention will be described more comprehensively and meticulously below in conjunction with the accompanying drawings and preferred embodiments of the present invention, but the protection scope of the present invention is not limited to the following specific embodiments.
[0033] Unless otherwise defined, all professional terms used below have the same meanings as those generally understood by those skilled in the art. The professional terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the scope of protection of the present invention.
[0034] Unless otherwise specified, various raw materials, reagents, instruments and equipment used in the present invention can be purchased from the market or prepared by existing methods.
[0035] The radioactive tracer [18F]FPEB, the precursor can be purchased on the market, and after the precursor is purchased, it can be [18F] labeled by the following method: prepare fluorine-18 by proton bombardment in a cyclotron, and after obtaining the radioactive fluorine-18, react it with the precursor compound, and after purification and quality control, obtain the radioactive tracer [18F]FPEB with a radiochemical purity of more than 98%.
[0036] Glossary:
[0037] mGluR5: metabotropic glutamate receptor 5;
[0038] PET: positron emission tomography;
[0039] MRI: magnetic resonance imaging;
[0040] TOF: Time of flight.
[0041] Example:
[0042] It should be understood that the method for constructing a depression prognosis prediction model provided in the present application can be applied to the scenario of identifying PET scan images and MRI images of the prognosis of depression patients. This is only an example and not a limitation.
[0043] A method for constructing a depression prognosis prediction model, such as Figure 1 As shown, the following steps are included:
[0044] (1) Acquisition of PET images and T1-weighted MRI images: To overcome the stability problem of mGluR5 receptor measurement, the present invention intravenously injected participants with approximately 195MBq [18F]FPEB (radiochemical purity exceeded 98%), and then collected equilibrium emission data on a high-resolution PET / CT scanner (Biograph mCTx, Siemens, Erlangen, Germany). The PET data were reconstructed using Siemens' TOF (Time-of-Flight) algorithm (iteration 8, image size 512, Gaussian filter, full width half maximum 1.0mm). Using PET data collected from 90 to 120 minutes after injection, it was shown that 18F-FPEB reached equilibrium during this period. At this stage, the whole brain cerebellum was used as a reference region to generate parametric images of the non-displacement binding potential (BPND) of the mGluR5 receptor. This dose plus continuous infusion mode was used to balance the modeled VT and BPND parameters, which had excellent test-retest reproducibility for the parameters of 18F-FPEB, ensuring the reliability and accuracy of the data. T1-weighted MRI scans were acquired on a 3T MRI scanner (uMR 790, United Imaging, Shanghai, China) using a 32-channel head coil to assess potential structural abnormalities and assist in the registration of PET data to standard space. The parameters were set as follows: TR = 6.97 ms, TE = 3.1 ms, flip angle = 10 degrees, slice thickness = 1 mm, slice interval = 0 mm, FOV = 230 × 256 mm 2 , matrix size = 460×512, number of slices = 200.
[0045] (2) Image preprocessing: The preprocessing of PET images was performed using the FMRIB software library (FSL, version 5.0.7) and SPM12, which consisted of the following steps:
[0046] 1) Using precise image registration technology, each participant's PET image is registered with the corresponding T1-weighted MRI scan image. This process first removes noise and corrects image deviations from the two different modalities to improve the accuracy and reliability of subsequent registration. The registration algorithm is based on the principle of mutual information and determines the optimal registration parameters by maximizing the amount of common information between PET and MRI images. Manual quality control is used to adjust and optimize the registration results to ensure that the anatomical structures between different images can be accurately aligned.
[0047] 2) Segmentation of T1-weighted MRI images using statistical parametric mapping (SPM) software. The purpose of this step is to finely divide brain tissue into different components such as gray matter, white matter, and cerebrospinal fluid. Through this segmentation, we can obtain gray matter and white matter probability masks representing different brain tissues, which provide important basic information for subsequent analysis. Subsequently, the Advanced Normalization Tool (ANTS) software package is used to spatially normalize the T1 images so that individual brain images can be registered to a standard brain space (MNI152 space). This step involves the comprehensive application of rigid (Rigid), affine (Affine), and deformable (Non-linear) transformations, with the goal of achieving accurate registration by minimizing the difference between the individual brain anatomical structure and the standard brain template. Rigid transformation is mainly used to correct image translation and rotation; affine transformation further adjusts image scaling, shearing, etc.; finally, deformable transformation can fine-tune the shape of individual brain regions to match the standard template to the greatest extent. After the T1 image is normalized, the resulting transformation parameters (transformation matrix) are applied to the corresponding PET image. Since the PET images and T1-weighted MRI images have been aligned through precise registration, this step can ensure that the functional information in the PET images corresponds accurately to the anatomical information and deform it into the unified MNI152 standard space. After the registration is completed, the accuracy of the registration is verified by visual inspection and automated quality assessment steps based on MRICRON and MATLAB to ensure that the image data used for subsequent analysis has a high degree of consistency and reliability.
[0048] 3) The simplified reference tissue model (SRTM) was used to calculate the non-displacement binding potential (BPND) of mGluR5 availability using the cerebellum as the reference region. The BPND values obtained by this method were highly correlated with the equilibrium analysis (EQ) results using arterial (EQA) or venous input data (EQV) and had good repeated measurement variability with minimal differences between different analysis methods.
[0049] (3) Model construction: After preprocessing, the availability of mGLUR5 receptors in each brain region was extracted using the AAL116 brain atlas. The t-test statistical test was used to perform univariate analysis on each feature to screen out features that were significantly associated with the prognosis of depression. Based on the preliminary t-test screening, the recursive feature elimination technique was used to select features more finely. The RFE technique optimizes the feature set by building a model and gradually removing the features that contribute the least to the prediction target. This process is repeated until the preset number of features or model performance threshold is reached. The key advantage of this method is that it can effectively reduce the data dimension while retaining the most critical biomarker information for predicting the prognosis of depression. The support vector machine (SVM) algorithm was used to build a preliminary model, and the SVM model was initialized using the Python scikit-learn machine learning library. The advantage of SVM is that it can handle data sets with complex decision boundaries. The kernel function is used to map the input space to a high-dimensional feature space. The most commonly used ones include linear kernels, polynomial kernels, and radial basis function kernels (RBF). Before training the SVM model, feature scaling was performed to avoid adverse effects on model performance due to dimensional differences in features. We used a standardization method, which scaled all features to a mean of 0 and a variance of 1 to ensure that the model had a uniform evaluation standard in all dimensions. To find the best combination of model parameters, a grid search strategy was used, which is a method for optimizing model performance by traversing a given parameter grid. In particular, we focused on three key parameters: C (regularization parameter), kernel (kernel function type), and gamma (kernel coefficient). The leave-one-out cross-validation (LOOCV) method was used to evaluate the model performance under different parameter combinations. In each leave-one-out cross-validation iteration, the accuracy, recall, precision, and F1 score of the model were recorded and calculated. The receiver operating characteristic curve (ROC) and area under the curve (AUC) of the model were calculated to evaluate the overall performance of the model. The performance indicators under different parameter settings were analyzed to select the model with the best comprehensive effect. The performance of the model in different patient groups was analyzed to ensure that the model had a good predictive effect in all types of patients. Analyze the prediction results of the model, explain which features are most critical for prognosis prediction, and ensure that the interpretability of the model is consistent with clinical knowledge to facilitate clinical decision-making. Explore the feasibility of the model in actual clinical settings, including ease of operation, cost-effectiveness, and potential clinical impact.
[0050] like Figure 2As shown, the accuracy of the model constructed by the present invention is 0.88, the recall rate is 0.93, the precision is 0.87, and the F1 score is 0.90. It can be seen that the depression prognosis prediction model based on mGluR5 of the present invention has good performance. The optimal model parameters obtained after grid search and parameter tuning are: 'C': 10, 'gamma': 0.1, 'kernel': 'linear'.
[0051] The present application also provides a system for constructing a depression prognosis prediction model, comprising:
[0052] A data acquisition module, used to acquire mGLUR5 receptor availability PET images of different patients through PET scanning, and then acquire T1-weighted MRI images;
[0053] An image preprocessing module to register the mGLUR5 receptor availability PET images of different patients with the corresponding T1-weighted MRI images, segment the T1-weighted MRI images using statistical parameter mapping software, and calculate the non-displacement binding potential of mGluR5 availability using a simplified reference tissue model with the cerebellum as the reference region;
[0054] A model building module is used to extract the mGLUR5 receptor availability of each brain region through the AAL116 brain atlas, use RFE technology to build a model and gradually remove the features that contribute the least to the prediction target to optimize the feature set. This process is repeated until the preset number of features or model performance threshold is reached; a preliminary model is built using a support vector machine algorithm, the model is initialized, the model performance is analyzed and optimized, and the model with the best comprehensive effect is selected as the depression prognosis prediction model.
[0055] The above-mentioned system for constructing a depression prognosis prediction model can implement various embodiments of the above-mentioned method for constructing a depression prognosis prediction model and can achieve the same beneficial effects, which will not be described in detail here.
[0056] Furthermore, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the aforementioned construction method is implemented when the processor executes the program.
[0057] Furthermore, the present application also provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the aforementioned construction method.
[0058] Furthermore, the present application also provides an application of mGluR5 as a biomarker in constructing a depression prognosis prediction model. Using mGluR5 as a biomarker, mGluR5 receptor availability data is obtained through PET scanning, and a depression prognosis prediction model is constructed based on the mGluR5 receptor availability data. The construction method is as described above.
Claims
1. A method for constructing a depression prognosis prediction model, characterized in that: The steps include: (1) PCT images of mGLUR5 receptor availability in different patients were obtained by PET scanning, and then T1-weighted MRI images were obtained; (2) registering the mGLUR5 receptor availability PET images of different patients obtained in step (1) with the corresponding T1-weighted MRI images, segmenting the T1-weighted MRI images using statistical parameter mapping software, and calculating the non-displacement binding potential of mGluR5 availability using a simplified reference tissue model with the cerebellum as the reference region; (3) After the image preprocessing in step (2) is completed, the mGLUR5 receptor availability of each brain region is extracted through the AAL116 brain atlas, and the model is constructed using the RFE technology and the features that contribute the least to the prediction target are gradually removed to optimize the feature set. This process is repeated until the preset number of features or model performance threshold is reached; a preliminary model is constructed using the support vector machine algorithm, the model is initialized, the model performance is analyzed and optimized, and the model with the best comprehensive effect is selected as the depression prognosis prediction model.
2. The construction method according to claim 1, characterized in that: The process of step (1) specifically includes the following steps: the patient group is intravenously injected with the radiotracer [18F]FPEB, and then balanced emission data are collected on a PET / CT scanner. The obtained data are used to locate the position of the radiotracer [18F]FPEB in the body using a TOF algorithm, and the mGLUR5 receptor availability PET image is reconstructed, and then a T1-weighted MRI image is acquired using a 32-channel head coil on a 3T magnetic resonance scanner.
3. The construction method according to claim 1, characterized in that: In step (2), the registration process is as follows: first, the images of the two different modalities are denoised and image deviations are corrected, and the optimal registration parameters are determined by maximizing the common information between the mGLUR5 receptor availability PET image and the MRI image, and then manual quality control is performed to adjust and optimize the registration results to ensure that the anatomical structures between the different images can be accurately aligned; The process of segmenting T1-weighted MRI images using statistical parametric mapping software includes the following steps: segmenting T1-weighted MRI images using statistical parametric mapping software to obtain gray matter and white matter probabilistic masks representing different brain tissues; then, spatially normalizing the T1 images using the ANTS software package to achieve accurate registration by minimizing the differences between individual brain anatomy and the standard brain template; finally, fine-tuning the shapes of individual brain regions to maximally match the standard template; After the T1 image normalization was completed, the resulting transformation parameters were applied to the corresponding PET images; after the registration was completed, the accuracy of the registration was verified by visual inspection and automated quality assessment steps based on MRICRON and MATLAB.
4. The construction method according to claim 1, characterized in that: In step (3), after extracting the mGLUR5 receptor availability of each brain region through the AAL116 brain atlas, a univariate analysis was performed on each feature using the t-test statistical test to screen out features that are significantly associated with the prognosis of depression. Based on the preliminary t-test screening, a recursive feature elimination technique was used to select features more finely; a preliminary model was constructed using the support vector machine algorithm, and the model was initialized using Python's scikit-learn machine learning library; before training the support vector machine algorithm model, a standardized method was used for feature scaling, that is, all features were scaled to a mean of 0 and a variance of 1 to ensure that the model had a unified evaluation standard in all dimensions.
5. The construction method according to claim 1, characterized in that: In step (3), the process of analyzing and optimizing model performance is as follows: a leave-one-out cross-validation method is used to evaluate model performance under different parameter combinations. In each leave-one-out cross-validation iteration, the accuracy, recall, precision, and F1 score of the model are recorded and calculated. The receiver operating characteristic curve and the area under the curve of the model are calculated to evaluate the overall performance of the model. Then, the performance indicators under different parameter settings are analyzed, and the model with the best comprehensive effect is selected as the depression prognosis prediction model.
6. A system for constructing a depression prognosis prediction model, characterized in that: include: A data acquisition module, used to obtain mGLUR5 receptor availability PET images of different patients through PET scanning, and then obtain T1-weighted MRI images; An image preprocessing module to register the mGLUR5 receptor availability PET images of different patients with the corresponding T1-weighted MRI images, segment the T1-weighted MRI images using statistical parameter mapping software, and calculate the non-displacement binding potential of mGluR5 availability using a simplified reference tissue model with the cerebellum as the reference region; A model building module is used to extract the mGLUR5 receptor availability of each brain region through the AAL116 brain atlas, use RFE technology to build a model and gradually remove the features that contribute the least to the prediction target to optimize the feature set. This process is repeated until the preset number of features or model performance threshold is reached; a preliminary model is built using a support vector machine algorithm, the model is initialized, the model performance is analyzed and optimized, and the model with the best comprehensive effect is selected as the depression prognosis prediction model.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the construction method according to any one of claims 1 to 5 is implemented.
8. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the construction method described in any one of claims 1 to 5.
9. Application of mGluR5 as a biomarker in constructing a prognosis prediction model for depression.
10. The use according to claim 9, characterized in that: Using mGluR5 as a biomarker, mGluR5 receptor availability data is obtained through PET scanning, and a depression prognosis prediction model is constructed based on the mGluR5 receptor availability data.