A method for constructing a pain sensitivity prediction model based on human cerebrospinal resting-state functional magnetic resonance imaging

Abstract

The present invention discloses a method for constructing a pain sensitivity prediction model based on human cerebrospinal resting-state functional magnetic resonance imaging, inducing a behavioral experiment of pain for each participant, collecting synchronous joint images of the brain and spinal cord and converting the format; segmenting the synchronous structural images of the brain and spinal cord, segmenting the synchronous resting-state functional images of the brain and spinal cord after preliminary preprocessing, obtaining separate structural image data and separate resting-state functional image data of the brain and spinal cord, performing subsequent preprocessing respectively, and aligning the obtained structural images to standard templates respectively, and aligning the functional images to the structural images; partitioning the obtained brain data, dividing the spinal cord data into several regions, constructing a brain and spinal cord resting-state functional connection matrix, and establishing a prediction model. The present invention provides a set of standardized cerebrospinal synchronous resting-state functional imaging data processing procedures and a prediction model establishment procedure to ensure the consistency and repeatability of the data, and facilitate promotion and application in different research institutions and clinical environments.

Description

Technical Field

[0001] The present invention belongs to the technical field of neuroscience research, and specifically relates to a method for constructing a pain sensitivity prediction model based on human cerebrospinal resting-state functional magnetic resonance imaging. Background Art

[0002] As two important components of the central nervous system, the brain and spinal cord play a key role in regulating and controlling human sensation, movement and various physiological functions. Research on the function of the brain and spinal cord can not only provide a deep understanding of the operating mechanism of the normal nervous system, but also reveal the pathological basis of a variety of neurological diseases. With the deepening of neuroscience research, researchers are paying more and more attention to the synergistic role of the brain and spinal cord in the function of the central nervous system. The complex interaction between the brain and spinal cord plays a key role in regulating pain, motor control and autonomic nervous function. Past studies have mainly focused on brain imaging or spinal cord imaging alone, but lack of systematic research on synchronous imaging of the two. The simultaneous cortico-spinal fMRI technology allows researchers to observe the activities of the brain and spinal cord at the same time, so as to have a more comprehensive understanding of the function of the central nervous system.

[0003] In recent years, the application of machine learning methods in neuroscience has gradually increased. By processing and analyzing high-dimensional neuroimaging data, valuable features can be extracted from it for disease diagnosis, prediction of treatment effects, and personalized medicine. Machine learning algorithms, especially prediction models based on neuroimaging data, can automatically identify important patterns related to diseases or physiological characteristics, greatly improving the efficiency and accuracy of data analysis. However, due to the high dimensionality and many features of imaging data, directly using all voxels for analysis will lead to the dimensionality disaster problem. Therefore, reducing the data dimension and selecting important features are of great significance to improving the prediction performance of the model.

[0004] In the field of pain research, understanding an individual's pain sensitivity is crucial to developing effective pain management strategies. Traditional pain assessment methods rely mainly on self-reports, which are highly subjective and difficult to quantify. Resting-state functional magnetic resonance imaging (rs-fMRI) is widely used to explore the functional connectivity of the brain and spinal cord. By capturing spontaneous neural activity, resting-state fMRI can study the functional connectivity between different areas of the brain and spinal cord in a task-free state, providing important information for understanding the organization and function of the central nervous system. Summary of the invention

[0005] In order to achieve image quality control of resting-state fMRI of the spinal cord, eliminate the influence of physiological noise and other noise when acquiring signals, and obtain stable neural activity signals from the images, the present invention provides a method for constructing a pain sensitivity prediction model based on resting-state functional magnetic resonance imaging of the human cerebrospinal cord.

[0006] The specific scheme is as follows:

[0007] A method for constructing a pain sensitivity prediction model based on human cerebrospinal resting-state functional magnetic resonance imaging

[0008] Step 1: Induce pain in each participant through behavioral experiments, collect pain sensitivity data, and convert the collected brain and spinal cord synchronized joint images into NIFTI format;

[0009] Step 2, segmenting the converted synchronous structural image of the brain and spinal cord to obtain separate structural image data of the brain and spinal cord;

[0010] Step 3, pre-processing the converted synchronous resting-state functional images of the brain and spinal cord, and then performing segmentation to obtain separate resting-state functional image data of the brain and spinal cord;

[0011] Step 4, performing subsequent preprocessing on the obtained individual data of the brain and spinal cord, and registering the obtained structural images to the standard templates, and registering the obtained functional images to the structural images;

[0012] Step 5, partition the obtained brain data according to the probability template, divide the C1 to C7 segments in the spinal cord data into several regions according to the probabilistic gray matter atlas, and construct the brain and spinal cord resting state functional connectivity matrix;

[0013] Step 6: Build a machine learning pipeline based on the brain and spinal cord resting-state functional connectivity matrix to establish a prediction model.

[0014] Preferably, in step 2, the coordinates of the upper end of the synchronous structural image C1 of the brain and spinal cord are firstly recorded, and then the synchronous structural image of the brain and spinal cord is divided into two parts, the brain and the spinal cord.

[0015] Preferably, the method for preliminarily preprocessing the converted brain and spinal cord synchronous resting state functional images is:

[0016] Synchronize the scan time of acquiring brain and spinal cord magnetic resonance images with breathing and heartbeat signals, create trigger acquisition signal files, and model physiological noise;

[0017] Respiration and heartbeat data were modeled using low-order Fourier series to generate several noise regressors and remove global physiological noise from brain and spinal cord functional data.

[0018] Distortion correction of the X and Y dimensions of rs-fMRI images was performed using two oppositely phase-encoded synchronized B0 field images of the brain and spinal cord;

[0019] Independent component analysis was used to decompose each participant's 4D brain and spinal cord synchronized functional MRI images, and the remaining noise was manually excluded;

[0020] Perform slice time correction and de-spike processing on the denoised fMRI images;

[0021] The supraaxial coordinates of C1 in fMRI images were recorded and segmented into the brain and spinal cord.

[0022] Preferably, in step 4, the brain and spinal cord structure data are subsequently preprocessed, and the obtained structure images are registered to the standard templates respectively, and the specific method is:

[0023] For the 3D structural data of the brain, the skull is first stripped, and then the white matter, gray matter, and cerebrospinal fluid are segmented;

[0024] For the 3D structural data of the spinal cord, the spinal cord was first segmented automatically to obtain a binary mask of the spinal cord, and then manually corrected according to the segmentation quality;

[0025] Manual marking of intervertebral discs to straighten the spinal cord;

[0026] Individual structural images were aligned to the standard vertebral stage by affine transformation;

[0027] A non-rigid transformation was used to register the structural image to the spinal cord standard template.

[0028] Further preferably, in step 4, the resting-state functional image of the brain is registered to the structural image, and the specific method is:

[0029] The obtained resting-state functional images of the brain were subjected to head motion correction and independent component analysis to eliminate the signal components related to head motion.

[0030] Binarized masks of white matter and CSF were registered back to individual functional space and time series were extracted;

[0031] The functional images were smoothed using a 5-mm full-width half-maximum (FWHM) Gaussian kernel;

[0032] A high-pass filter with a 100-second cutoff period was applied to remove low-frequency signals;

[0033] Individual functional space data were aligned to individual structural space and rigid, affine, and nonlinear registration was performed to the standard MNI (2 mm) brain template.

[0034] Further preferably, in step 4, the spinal cord resting state functional image is registered to the structural image, and the specific method is:

[0035] Motion correction was applied, and a cylindrical mask along the mid-axis of the spinal cord was drawn to exclude areas outside the spinal cord;

[0036] Spatial smoothing was performed using a 2D Gaussian kernel along the x and y axes with a FWHM of 3 mm;

[0037] The top 10% of the voxel time series with the largest variance were calculated and regressed at each time point using the x and y translations generated by motion correction;

[0038] The spinal cord data were aligned from individual functional space to individual structural space and then to the PAM50 standard space.

[0039] Preferably, in step 5, the functional data of the brain and spinal cord are divided into different regions of interest (ROIs) using probability templates of the brain and spinal cord; the lower triangular matrix of the functional connectivity matrix is ​​extracted and vectorized after constructing the resting-state functional connectivity matrix of the brain and spinal cord using the partial correlation method.

[0040] Preferably, in step 6, the brain-spinal cord functional connectivity vector obtained in step 5 is used as a feature to construct a machine learning pipeline and a prediction model, and the specific method is:

[0041] Robust feature scaling was used to remove the median and scale the data according to the quantile range;

[0042] The K-best feature selection method was used to select important features related to pain threshold;

[0043] The Elastic Net regression model is used to build the prediction model;

[0044] A grid search algorithm with an internal validation criterion was used to optimize the free hyperparameters of the machine learning pipeline in the training set to determine the optimal hyperparameters and obtain the prediction model.

[0045] Preferably, the objective function of the adopted Elastic Net regression model is:

[0046]

[0047] Among them, n is the number of samples; α is the regularization weight; l1_ratio is the L1 / L2 regularization ratio;

[0048] y is the true value vector; X is the feature matrix containing the eigenvalues ​​of each sample;

[0049] w is the weight vector of the model; Xw is the predicted value calculated using the model weights.

[0050] The technical solution of the present invention has the following advantages:

[0051] A. The present invention provides a set of standardized processes for processing brain-spinal synchronized resting-state functional imaging data and establishing prediction models to ensure data consistency and repeatability; this standardized process can be used across magnetic resonance imaging sites, facilitating its promotion and application in different research institutions and clinical settings.

[0052] B. The prediction model constructed by the present invention combines the resting-state functional connectivity data of the cerebrospinal fluid and uses a machine learning algorithm to achieve efficient prediction of individual pain sensitivity. The functional connectivity matrix is ​​constructed using the partial correlation method to extract important features, thereby improving the prediction performance of the prediction model. The K-best feature selection method is used to screen important features related to the pain threshold, and the Elastic Net regression model is used for prediction. The model parameters are optimized through grid search and permutation test to ensure the robustness of the model and the prediction accuracy.

[0053] C. The present invention verifies the predictive ability and generalization performance of the model through multiple independent experiments, ensures the stability and applicability of the prediction model on different data sets, and improves the credibility and application prospects of the model.

[0054] D. The prediction model constructed by the present invention has important application value in clinical pain management and personalized treatment. By effectively predicting individual pain sensitivity, it can help doctors develop more accurate and personalized pain management strategies and improve the quality of life of patients. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] In order to more clearly illustrate the specific embodiments of the present invention, the drawings required for use in the specific embodiments 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.

[0056] Figure 1 is a flow chart of the method for constructing a pain sensitivity prediction model provided by the present invention;

[0057] Figure 2 It is a schematic diagram of the construction and prediction of the pain sensitivity prediction model provided by the present invention;

[0058] Figure 3 are 29 non-zero prediction connection weights included in the prediction model of the present invention;

[0059] Figure 4 It is an effective prediction of the thermal pain sensitivity of the contents of Study 2 and Study 3 provided by the present invention;

[0060] Figure 5 It is an effective prediction of the thermal pain sensitivity of the contents of Study 4 and Study 5 provided by the present invention. DETAILED DESCRIPTION

[0061] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0062] The present invention provides a method for constructing a pain sensitivity prediction model based on human cerebrospinal resting state functional magnetic resonance imaging. Figure 1 and Figure 2 As shown, the specific steps include:

[0063]

S01

[0064]

S02

[0065] [S03] The converted synchronous resting-state functional images of the brain and spinal cord are first preprocessed and then segmented to obtain separate resting-state functional image data of the brain and spinal cord.

[0066] For each participant's brain-spinal synchronized rs-fMRI images, the scan time was first synchronized with the respiratory and heartbeat signals to create a triggered acquisition signal file and model physiological noise. This modeling assumes that the physiological process is quasi-periodic, and the respiratory and heartbeat data are modeled by low-order Fourier series, generating a total of 32 noise regressors, and removing global physiological noise from the brain-spinal synchronized functional data; then, two opposite phase-encoded brain-spinal synchronized B0 field images were used to correct the distortion of the X and Y dimensions of the rs-fMRI images; then, the 4D brain-spinal synchronized functional MRI images of each participant were decomposed using independent component analysis, and the remaining noise was manually excluded; the denoised fMRI images were sliced ​​time corrected and de-spiked. Finally, the supraaxial coordinates of C1 in the fMRI images were recorded and segmented into the brain and spinal cord for subsequent preprocessing.

[0067] [S04] The obtained individual data of the brain and spinal cord are subsequently preprocessed, and the obtained structural images are registered to the standard templates, and the obtained functional images are registered to the structural images.

[0068] For the 3D structural data of the brain, the skull is first stripped and then the white matter, gray matter and cerebrospinal fluid are segmented.

[0069] For the 3D structural data of the spinal cord, the spinal cord was first segmented automatically to obtain a binary mask of the spinal cord, and then manually corrected according to the segmentation quality. Subsequently, the intervertebral discs were manually marked to straighten the spinal cord. The individual structural images were aligned to the standard vertebral stage by affine transformation, and then the images were registered to the spinal cord standard template (PAM50 T1 weighted template) by non-rigid transformation.

[0070] For the resting-state fMRI images of the brain, head motion correction and independent component analysis were first performed to eliminate signal components related to head motion; the binary masks of white matter and cerebrospinal fluid were aligned back to the individual functional space and the time series were extracted to eliminate the signal effects of white matter and cerebrospinal fluid in the functional image data. The functional image was further smoothed using a 5mm full-width half-maximum (FWHM) Gaussian kernel to enhance the signal-to-noise ratio. Then, a high-pass filter with a 100-second cutoff period was applied to remove low-frequency signals, thereby reducing the influence of non-neural noise and improving the temporal stability of the data. Finally, the individual functional space data was aligned to the individual structural space, and then rigid, affine, and nonlinear registration was performed to the standard MNI (2mm) brain template.

[0071] For resting-state fMRI images of the spinal cord, motion correction was first performed, and a cylindrical mask along the midline of the spinal cord was drawn to exclude areas outside the spinal cord. Next, spatial smoothing was performed using a 2D Gaussian kernel along the x and y axes with a FWHM of 3 mm. To account for noise in the cerebrospinal fluid, the top 10% of the voxel time series with the largest variance were calculated and regressed using the x and y translations generated by motion correction at each time point. Finally, the spinal cord data were registered from individual functional space to individual structural space and then to the PAM50 standard space.

[0072]

S05

[0073] [S06] Using the brain and spinal cord functional connectivity vector obtained in step [S05] as a feature to construct a machine learning pipeline to build a prediction model. The specific steps include:

[0074]

S061

[0075]

S062

[0076] [S063] Model construction: The Elastic Net regression model is used for prediction, which combines L1 and L2 regularization to adjust sparsity. The objective function of the Elastic Net regression model is as follows:

[0077]

[0078] Among them, n is the number of samples; α is the regularization weight; l1_ratio is the L1 / L2 regularization ratio;

[0079] y is the true value vector; X is the feature matrix containing the eigenvalues ​​of each sample;

[0080] w is the weight vector of the model; Xw is the predicted value calculated using the model weights.

[0081]

S064

[0082]

S065

[0083] Example:

[0084] Study 1: Fifty-four healthy subjects (31 females, average age 22.3 years) were randomly recruited to participate in a brain and spinal cord simultaneous imaging magnetic resonance imaging experiment;

[0085] Study 2: Two weeks later, 35 participants (20 females, average age 22.1 years) from Study 1 were recruited again as a validation set for the model.

[0086] Study 3: 40 healthy subjects (16 females, average age 21.4 years) were randomly recruited.

[0087] All subjects were right-handed, healthy, had no history of chronic pain or mental illness, and had not taken analgesics recently.

[0088] During the experiment, the subjects were arranged to participate in a behavioral experiment of temperature-induced pain before the combined brain and spinal cord scanning. The thermal pain stimulus was placed on the right arm of the subject through a hot electrode, 10 cm away from the wrist. The baseline temperature was set at 32°C, and the thermal threshold was obtained by gradually increasing the temperature stimulus (1°C / s). The stimulus was terminated when the participant pressed a button. This procedure for assessing thermal pain sensitivity was repeated five times and the average was taken to ensure the accuracy and consistency of the measurement, thereby providing an overall assessment of pain sensitivity.

[0089] Study 4: 14 patients with diabetes and chronic pain (5 females, average age 59.83) were randomly recruited from Shandong Provincial Hospital. All patients underwent brain-spinal synchronous imaging magnetic resonance imaging and filled out the McGill Pain Questionnaire-Short Form.

[0090] Study 5: included 20 patients with postherpetic neuralgia (PHN) from Beijing Xuanwu Hospital (3 females, average age 55.22 years old), who recorded the current pain score using a 0-10 numerical rating scale (NRS).

[0091] Combined with the above steps [S01] and [S04], the experimental data collected synchronously from the brain and spinal cord are preprocessed, and then subsequent statistical analysis is performed and a machine learning prediction model is constructed. The specific steps are detailed as follows [S05] to [S08].

[0092]

S05

[0093] [S06] Use the 9591 vectors of brain-spinal cord functional connectivity obtained in [S05] as features to build a machine learning pipeline to establish a prediction model. The specific steps include:

[0094]

S061

[0095]

S062

[0096]

S063

[0097]

S064

[0098] [S065] To evaluate whether the predictive ability of the model exceeds random probability and to ensure the fairness of cross-validation, the present invention uses a one-sided permutation test to evaluate the significance of the model. The permutation test is performed by randomly shuffling the thermal pain threshold labels of the participants for 10,000 iterations. Finally, the prediction model (K = 2000, l1_ratio = 0.999, α = 0.1) was named Corticospinal Pain Sensitivity Signature (CSps).

[0099] [S066] To evaluate the stability of the predicted non-zero weighted functional connections, the present invention performed a bootstrap test. 10,000 resampled samples, each with the same size as the original dataset, were generated by randomly selecting and replacing observations in the original data. Subsequently, the prediction model was fitted using the optimal hyperparameters, and the uncorrected p-value for each selected functional connection was calculated. Figure 3 As shown, CSps includes 29 non-zero predictive connection weights, including 16 positive weights and 13 negative weights. Regions with positive (negative) predictive weights indicate that higher functional connectivity predicts higher thermal pain thresholds. Nearly half of the variance in predicted pain sensitivity scores is explained by the six strongest functional connections described below.

[0100] The most significant positive functional connections (predicting greater pain sensitivity) were: (1) the anteromedial somatomotor network (M1 and supplementary motor area) to the ventral horn of the spinal cord C4, representing a direct connection between the brain sensorimotor network and spinal cord regions that integrate sensorimotor processing; and (2) the right dorsolateral somatomotor network to the left posterior middle frontal gyrus.

[0101] The most significant negative functional connections (predicting lower pain sensitivity) were: (1) ventral precuneus to left frontal pole; (2) hippocampus to lobule IV of cerebellar hemisphere; (3) posterior cingulate cortex to rostral part of inferior frontal gyrus; and (4) hippocampus to vermis of cerebellar hemisphere.

[0102]

S07

[0103] In Study 2, the CSps model performed well in predicting heat pain threshold scores, with a Pearson correlation coefficient of 0.52 (p = 0.0003, one-sided permutation test), an explained variance of 27%, and a mean absolute error (MAE) of 0.62 (p = 0.0007, one-sided permutation test).

[0104] In Study 3, the CSps model still performed well, with a Pearson correlation coefficient of 0.45 (p = 0.003, one-sided permutation test), an explained variance of 18%, and a mean absolute error (MAE) of 0.63 (p = 0.007, one-sided permutation test).

[0105]

S08

[0106] In summary, the pain sensitivity prediction model based on resting-state functional connectivity data between the brain and spinal cord provided by the present invention can effectively predict the pain sensitivity of an individual and is suitable for basic research and clinical applications.

[0107] Obviously, the above embodiments are merely examples for the purpose of clear explanation, and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived therefrom are still within the scope of protection of the present invention.

Claims

1. A method for constructing a pain sensitivity prediction model based on human cerebrospinal resting state functional magnetic resonance imaging, characterized in that: Step 1: Induce pain in each participant, collect pain sensitivity data, and convert the collected brain and spinal cord synchronized joint images into NIFTI format; Step 2, segmenting the converted synchronous structural image of the brain and spinal cord to obtain separate structural image data of the brain and spinal cord; Step 3, pre-processing the converted synchronous resting-state functional images of the brain and spinal cord, and then performing segmentation to obtain separate resting-state functional image data of the brain and spinal cord; Step 4, performing subsequent preprocessing on the individual data of the brain and spinal cord obtained in step 2 and step 3, and registering the obtained structural images to the standard templates, and registering the obtained functional images to the structural images; registering the resting state functional images of the brain to the structural images, the specific method is: The obtained resting-state functional images of the brain were subjected to head motion correction and independent component analysis to eliminate the signal components related to head motion. Binarized masks of white matter and CSF were registered back to individual functional space and time series were extracted; The functional image was smoothed using a 5 mm full width half maximum (FWHM) Gaussian kernel; A high-pass filter with a 100-second cutoff period was applied to remove low-frequency signals; Align individual functional space data to individual structural space, and perform rigid, affine, and nonlinear registration to the standard MNI-2mm brain template; The specific method for registering the resting-state functional image of the spinal cord to the structural image is: Motion correction was applied, and a cylindrical mask along the mid-axis of the spinal cord was drawn to exclude areas outside the spinal cord; Spatial smoothing was performed using a 2D Gaussian kernel along the x and y axes with a FWHM of 3 mm; The top 10% of the voxel time series with the largest variance were calculated and regressed at each time point using the x and y translations generated by motion correction; The spinal cord data were aligned from individual functional space to individual structural space, and then aligned to the PAM50 standard space; Step 5, partition the obtained brain data according to the probability template, divide the C1 to C7 segments in the spinal cord data into several regions according to the probabilistic gray matter atlas, and construct the brain and spinal cord resting state functional connectivity matrix; Step 6: construct a machine learning pipeline based on the brain and spinal cord resting state functional connectivity matrix to establish a prediction model. The specific method is: Robust feature scaling was used to remove the median and scale the data according to the quantile range; The K-best feature selection method was used to select important features related to pain threshold; The Elastic Net regression model is used to build the prediction model; A grid search algorithm with an internal validation criterion was used to optimize the free hyperparameters of the machine learning pipeline in the training set to determine the optimal hyperparameters and obtain the prediction model.

2. The method for constructing a pain sensitivity prediction model based on human cerebrospinal resting state functional magnetic resonance imaging according to claim 1, characterized in that: In the step 2, the coordinates of the upper end of the synchronous structural image C1 of the brain and spinal cord are firstly recorded, and then the synchronous structural image of the brain and spinal cord is divided into two parts, the brain and the spinal cord.

3. The method for constructing a pain sensitivity prediction model based on human cerebrospinal resting state functional magnetic resonance imaging according to claim 1, characterized in that: In step 3, the method for preliminarily preprocessing the converted synchronous resting-state functional image of the brain and spinal cord is: Synchronize the scan time of acquiring brain and spinal cord magnetic resonance images with breathing and heartbeat signals, create trigger acquisition signal files, and model physiological noise; Respiration and heartbeat data were modeled using low-order Fourier series to generate several noise regressors and remove global physiological noise from brain and spinal cord functional data. Distortion correction of the X and Y dimensions of rs-fMRI images was performed using two oppositely phase-encoded synchronized B0 field images of the brain and spinal cord; The 4D brain and spinal cord synchronized functional MRI images of each participant were decomposed using independent component analysis, and the remaining noise was manually excluded; The denoised fMRI images were subjected to slice time correction and spike removal; The supraaxial coordinates of C1 in fMRI images were recorded and segmented into the brain and spinal cord.

4. The method for constructing a pain sensitivity prediction model based on human cerebrospinal resting state functional magnetic resonance imaging according to claim 1, characterized in that: In step 4, the brain and spinal cord structure data are subsequently preprocessed, and the obtained structure images are registered to the standard templates respectively, and the specific method is: For the 3D structural data of the brain, the skull is first stripped, and then the white matter, gray matter, and cerebrospinal fluid are segmented; For the 3D structural data of the spinal cord, the spinal cord was first segmented automatically to obtain a binary mask of the spinal cord, and then manually corrected according to the segmentation quality; Manual marking of intervertebral discs to straighten the spinal cord; Individual structural images were aligned to the standard vertebral stage by affine transformation; A non-rigid transformation was used to register the structural image to the spinal cord standard template.

5. The method for constructing a pain sensitivity prediction model based on human cerebrospinal resting state functional magnetic resonance imaging according to claim 1, characterized in that: In step 5, the functional data of the brain and spinal cord are divided into different regions of interest (ROIs) using probability templates of the brain and spinal cord; the resting-state functional connectivity matrix of the brain and spinal cord is constructed using the partial correlation method, and then the lower triangular matrix of the functional connectivity matrix is ​​extracted and vectorized.

6. The method for constructing a pain sensitivity prediction model based on human cerebrospinal resting state functional magnetic resonance imaging according to claim 1, characterized in that: The objective function of the Elastic Net regression model used is: ； Where n is the number of samples; α is the regularization weight; l1_ratio is the L1 / L2 regularization ratio; y is the true value vector; X is the feature matrix containing the eigenvalues ​​of each sample; w is the weight vector of the model; Xw is the predicted value calculated using the model weights.

Images