Pain intensity model construction method based on task-state brain-spine synchronous functional magnetic resonance

Abstract

The invention discloses a pain intensity prediction model construction method based on task state brain-spine synchronous functional magnetic resonance, which comprises the following steps: inducing a pain stimulation task of each subject, collecting brain and neck marrow synchronous task state functional image data, and converting the data into an NIFTI format; segmenting the synchronous structure image to obtain independent structure image data of the brain and the neck marrow, and performing segmentation after preprocessing to obtain independent task state function image data of the brain and the neck marrow; performing subsequent preprocessing on the obtained independent data of the brain and the neck marrow, registering the structure image of the data to a standard template, and registering the task state function image of the data to the structure image; performing first-order statistical analysis on the obtained task state function image to generate a comparison parameter diagram of each subject; and constructing a machine learning pipeline, and developing and verifying a prediction model to obtain a pain intensity prediction model. The pain intensity prediction model (CsPIP) constructed by the invention directly decodes transient nerve response to pain stimulation, and can reflect subjective pain intensity more directly and dynamically.

Description

Technical Field

[0001] This invention belongs to the field of interdisciplinary technology of neuroimaging and artificial intelligence. More specifically, it relates to a method for constructing a pain intensity prediction model based on task-oriented brain and spinal cord synchronous functional magnetic resonance imaging, which is used for tracking and objectively evaluating the efficacy of analgesic interventions. Background Technology

[0002] Pain is a complex subjective experience, and its objective quantification has always been a challenge in clinical practice and scientific research. Traditional pain assessment mainly relies on patient self-reporting, which is highly subjective and easily influenced by various factors. In recent years, neuroimaging has been dedicated to finding objective pain biomarkers. Brain central models, represented by neuropain signatures (NPS), based on task-based functional magnetic resonance imaging (fMRI) and using multi-voxel pattern analysis (MVPA), have made significant progress in predicting pain intensity at the cortical level. However, these models have a fundamental flaw: they completely ignore the spinal cord's role as a key hub for the primary integration and gating of pain information, failing to encompass the complete neural axis mechanisms of pain processing, thus limiting their predictive accuracy and mechanistic completeness.

[0003] To compensate for the shortcomings of pure brain models, existing technologies have developed pain sensitivity prediction models based on resting-state brain-spinal functional connectivity. These models predict an individual's pain threshold by analyzing the correlation between spontaneous activity of the brain and spinal cord in a task-free state. However, resting-state functional connectivity reflects the inherent organizational patterns of neural networks, rather than a direct response to specific stimuli (such as pain), and therefore is not suitable for neural coding that directly and dynamically reflects the intensity of instantaneous pain.

[0004] Task-based functional magnetic resonance imaging (fMRI) directly captures changes in blood oxygen level-dependent (BOLD) signals in the brain and spinal cord as they respond to standardized noxious stimuli (such as heat pain). This paradigm can more directly reveal neural activity patterns directly related to the intensity of pain perception. Current technology lacks a standardized scheme and predictive model specifically based on task-based simultaneous brain and spinal cord fMRI that uses multivariate pattern analysis to directly decode subjective pain intensity. Summary of the Invention

[0005] To address the aforementioned technical problems in existing technologies, this invention provides a method for constructing a pain intensity model based on task-oriented simultaneous functional magnetic resonance imaging (fMRI) of the brain and cervical spinal cord. For the first time, it utilizes simultaneously acquired brain and cervical spinal cord fMRI to establish a pain intensity prediction model under a pain stimulus task using multivariate methods such as LASSO-PCR.

[0006] The specific plan adopted is as follows: A method for constructing a pain intensity prediction model based on task-based simultaneous functional magnetic resonance imaging (fMRI) of the brain and spinal cord includes the following steps: Step 1: Induce each subject to a pain stimulus task, collect brain and cervical spinal cord synchronous task-state functional imaging data on the magnetic resonance platform, and convert them into NIFTI format; Step 2: Segment the converted brain and cervical spinal cord synchronous structural images to obtain separate structural image data of the brain and cervical spinal cord; Step 3: After preprocessing the synchronous task-mode functional images of the brain and cervical spinal cord, segmentation is performed to obtain separate task-mode functional image data of the brain and cervical spinal cord. Step 4: Perform subsequent preprocessing on the individual brain and cervical spinal cord data obtained in Step 2 and Step 3, and register their structural images to standard templates and their task-state functional images to structural images. Step 5: Perform first-order statistical analysis on the task-state functional images obtained in Step 4 to generate a comparative parameter map for each subject under pain stimulus conditions. Step 6: Use the comparison parameter map as the input features of the model to build a machine learning pipeline, and develop and validate the prediction model to obtain the pain intensity prediction model.

[0007] Specifically, the method for developing a pain intensity prediction model in step 6, which uses the contrast parameter map as the model input feature to construct a machine learning pipeline, is as follows: Step 6.1: Standardize the comparison parameter graph; Step 6.2: Use the LASSO-PCR regression model to predict pain intensity; Step 6.3: Use cross-validation to optimize the hyperparameters of the machine learning pipeline in the training set, determine the optimal hyperparameters, and obtain the pain intensity prediction model; Step 6.4: Perform a generalization test on the pain intensity prediction model on an independent test dataset to evaluate the significance of the pain intensity prediction model.

[0008] Preferably, the brain-cervical spinal cord synchronous magnetic resonance images of each subject are scanned using a 3.0T magnetic resonance imaging system, and equipped with a head-neck combined multichannel coil that simultaneously covers the brain, brainstem and cervical spinal cord.

[0009] Preferably, the task-state functional imaging employs a multi-echo planar imaging sequence that simultaneously covers the brain and cervical spinal cord, with a field of view of 192mm × 192mm. Seventy axial slices are arranged along the head-to-tail direction, with a single slice thickness of 4mm, so that the scanning volume extends from the medulla oblongata and the whole brain to the C1–C7 segments of the cervical spinal cord.

[0010] Preferably, the spatial resolution of the task-state functional image is a planar resolution of 1.5mm×1.5mm, the repetition time (TR) is 2.6 to 2.8s, the echo time (TE) is 25 to 30 ms, the flip angle is 70 to 90°, the phase encoding direction adopts a foot-head axis front-back configuration, and it is combined with parallel imaging and multi-band excitation to cover the entire axis of the brain and cervical spinal cord.

[0011] Furthermore, before or during the task-oriented functional image scan, at least one set of B0 field images with the same geometric parameters as the functional images but opposite phase encoding directions are acquired, and the subjects' breathing and heartbeat signals are recorded simultaneously.

[0012] Preferably, the synchronous acquisition of structural images of the brain and cervical spinal cord uses a three-dimensional T1-weighted sequence with an isotropic resolution of 1 mm³, a TR / TI of 2300 / 900 ms, and a TE of 2–3 ms.

[0013] In step 2, the coordinates of the upper end of the brain-spinal cord synchronous structural image C1 are first recorded, and then the brain-spinal cord synchronous structural image is divided into two parts: the brain and the spinal cord.

[0014] In step 3, the method for preprocessing the brain and cervical spinal cord synchronized task-state functional images is as follows: Synchronize the scan time with respiratory and heartbeat signals to create a trigger acquisition signal file; The respiratory and heart rate data were modeled using low-order Fourier series to generate several noise regressions, and global physiological noise was removed from the cortical spinal cord function data. Distortion correction of the X and Y dimensions of fMRI images was performed using two cortical spinal cord B0 field images encoded with opposite phases.

[0015] In step 4, the obtained individual brain and cervical spinal cord data are preprocessed separately, and their structural images are registered to a standard template. The specific method is as follows: For the 3D structural data of the brain, the skull is first dissected, and then the white matter, gray matter and cerebrospinal fluid are segmented. For the 3D structural data of the spinal cord, the spinal cord is first automatically segmented to obtain a binary mask of the spinal cord, and then manually corrected according to the segmentation quality. Manually mark the intervertebral discs to straighten the spinal cord; The individual structural image is aligned to the standard vertebral stage through affine transformation; Non-rigid transformations are used to register structural images to a standard spinal cord template.

[0016] In step 4, the task-oriented functional image of the brain is registered to the structural image. The specific method is as follows: Head movement correction was performed on the obtained task-oriented functional images of the brain. The binary mask of white matter and cerebrospinal fluid was registered back to the individual functional space and the time series was extracted; The functional images were smoothed using a 5mm full-width half-maximum (FWHM) Gaussian kernel; Low-frequency signals are removed by applying a high-pass filter with a 100-second truncation period; Individual functional space data are aligned to individual structural space and rigidly, affinely, and nonlinearly registered to a standard MNI (2 mm) brain template.

[0017] In step 4, the specific method for registering the spinal cord task-oriented functional image to the structural image is as follows: Motion correction was employed, and a cylindrical mask was drawn along the spinal cord axis to exclude areas outside the spinal cord. Spatial smoothing was performed along the x and y axes using a 2D Gaussian kernel with an FWHM of 3 mm. Calculate the top 10% of variances in the voxel time series, and perform regression at each time point using the x and y shifts generated by motion correction; The spinal cord data was registered from the individual functional space to the individual structural space, and then registered to the PAM50 standard space.

[0018] The technical solution of the present invention has the following advantages: A. This invention is the first to construct a pain intensity prediction model (CsPIP) based on task-based simultaneous brain and spinal cord MRI, which directly decodes the instantaneous neural response to pain stimuli and can more directly and dynamically reflect the subjective pain intensity.

[0019] B. The method of this invention integrates signals from the spinal cord (especially pain coding in the dorsal horn and inhibitory signals in the ventral horn) with the brain's classical pain matrix, and the prediction accuracy of the pain intensity prediction model is significantly better than models based solely on the brain (such as NPS) or models based solely on the spinal cord.

[0020] C. The CsPIP model constructed in this invention is ineffective against non-painful somatosensory stimuli (itching) and observational pain, but shows good specificity for nociceptive pain. At the same time, the model can sensitively capture the reduction in pain intensity brought about by analgesic interventions (such as TENS), and can be used as an objective biomarker for efficacy evaluation.

[0021] D. The CsPIP model constructed in this invention provides a new tool with clear translational potential for the objective assessment of pain. Its clinical applications are specifically reflected in: 1) providing objective and quantitative biomarkers for analgesic efficacy in drug and device clinical trials; 2) providing a basis for target verification and efficacy evaluation of neuromodulation therapies (such as TENS and transcranial magnetic stimulation); and 3) assisting in the objective diagnosis and precise classification of chronic pain.

[0022] E. This invention reveals the unique value and underlying mechanisms of the CsPIP model in patients with chronic pain: such as Figure 5 As shown, this model can not only quantitatively track therapeutic pain relief from patients' resting-state brain activity, but also... Figure 5 c) Furthermore, by defining the corticospinal dynamic state, it can identify the alternating "pro-pain" and "anti-pain" states in the patient's brain ( Figure 5 This discovery provides a novel explanation for understanding the variability of chronic pain and lays the theoretical foundation for developing state-dependent personalized neuromodulation therapies (such as interventions during pro-pain states), demonstrating great potential for clinical translation. Attached Figure Description

[0023] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart of the pain intensity prediction model construction method provided by the present invention; Figure 2 The experimental design and brain-spinal activation diagram provided for this invention: (a) A flowchart of the pain paradigm in Study 1, showing the sequential fixation point, thermal stimulation, rating, and trial interval.

[0025] (b) The block design of Study 2 showed pain or itching conditions of varying intensities (sequentially balanced), with self-reported scores following each condition.

[0026] (c) Pain-induced activation across the entire neural axis, group-level statistical parametric plots of heat-induced activation from Dataset 1, showing the distributed response across cortical, subcortical, and cervical spinal cord regions. Sagittal and transverse views of the brain and spinal cord illustrate regions with significantly higher pain activation than baseline (model settings and thresholds are described in the Methods section). Color bars represent the Z-values ​​and 1-p values ​​used for display.

[0027] Figure 3 The weight distribution diagram of the CsPIP model provided by this invention is as follows: Voxel CsPIP weights displayed on cortical, subcortical, cerebellar, and cervical spinal cord (C1–C7) sections; Warm colors indicate areas where higher activity levels predict higher pain intensity (e.g., anterior insula aINS, posterior insula pINS, midinsula midINS, dorsal anterior cingulate cortex / supplementary motor area dACC / SMA, secondary somatosensory cortex S2, tectal / posterior parietal cortex PT, thalamus THAL). Cool tones highlight negatively weighted areas (e.g., precuneus, frontoparietal / midfrontal gyrus FP / MFG). This visualization uses a thresholded weight map for display purposes only; the predictive model uses the full, unthresholded weight map.

[0028] Figure 4 The clinical sensitivity verification diagram of the model provided by this invention: The TENS experimental procedure, group-level behavioral results, and the coupling relationship between CsPIP feature differences and changes in pain scores are provided. Among them: (a) Workflow of Study 1. Ninety-two healthy participants first completed a thermal pain rating task (pre-intervention). After a short rest, they received active TENS or sham stimulation and then repeated the same pain task (post-intervention). Sixty-one participants completed the post-intervention phase and constituted the Study 1-TENS sample used in the analysis.

[0029] (b) Sham stimulus condition. The bar chart shows the mean ± standard error of subjective pain scores before and after sham stimulation when stimulation was applied to the left and right forearms; black dots represent individual participants. The horizontal line above the bar chart indicates the paired intra-group comparisons used.

[0030] (c) Sham stimulation conditions: Scatter plots of changes in CsPIP characteristic responses (x-axis) and pain scores (y-axis) before and after stimulation of the left and right arms. Shaded areas represent 95% confidence intervals.

[0031] (d) Active TENS conditions, plotted in the same way as (b).

[0032] (e) Active TENS conditions, plotted in the same way as (c).

[0033] Figure 5 The diagram showing the expression of CsPIP and cortical spinal cord dynamics in patients with chronic pain as revealed in this invention: (a) Displays the correlation matrix showing the relationships between demographic variables, clinical scales, and characteristic responses in the IBS cohort. The squares represent Pearson r values ​​(colored by size and positive / negative), and the connecting arcs highlight the significant associations between CsPIP and NPS.

[0034] (b) Violin plot of chronic pain scores before and after electroacupuncture treatment (each line represents one patient). The pain in the group was significantly reduced after treatment.

[0035] (c) The scatter plot shows that the change in CsPIP expression (ΔCsPIP) based on the resting-state fALFF profile is positively correlated with the change in pain score (Δpain) (r = 0.46), indicating that CsPIP can track treatment-induced pain relief. Each point represents one patient; the lines and shaded bands show the regression fit line and 95% confidence interval.

[0036] (d) Schematic diagram of a hidden Markov model applied to the corticospinal voxels defined by CsPIP, which divides the resting time series into four cyclic states.

[0037] (e) Model selection curve, plotted as a function of the cross-validation log-likelihood value as a function of the number of states; K = 4 (red line) was selected as the optimal solution.

[0038] (f) Comparison of pain values ​​predicted by CsPIP with baseline pain scores in each state highlights a nociceptive state (state 1, where CsPIP predicted values ​​are higher when pain is higher) and a resilient state (state 3, where the pattern is reversed).

[0039] (g) An example of a decoded state sequence of a representative subject, showing the rapid switching of four states over time (TRs).

[0040] (h) Stacked bar chart of the percentage of time spent in each state across patients; each column represents a subject, and the color indicates the percentage of time spent in each state. Detailed Implementation

[0041] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] like Figure 1 As shown, this invention provides a method for constructing a pain intensity prediction model based on task-oriented simultaneous functional magnetic resonance imaging (fMRI), comprising the following steps:

S01

[0043] The brain and cervical spinal cord synchronous magnetic resonance images of each subject were obtained using a 3.0T magnetic resonance imaging system equipped with a head-neck combined multichannel coil (e.g., a 64-channel head-neck coil) that simultaneously covers the brain, brainstem, and cervical spinal cord. Subjects were placed supine on the scanning table, with their head and neck immobilized by foam pads on both sides of the head and neck, a pillow for elevation, and a cervical collar to restrict head and neck movement. During the scan, subjects were instructed to avoid swallowing and large movements as much as possible to reduce motion artifacts and ensure synchronous and stable acquisition of brain and spinal cord signals.

[0044] The task-oriented functional imaging employs a multi-echo planar imaging sequence that simultaneously covers the brain and cervical spinal cord. The preferred field of view is 192mm × 192mm, with 70 axial slices arranged along the head-to-tail direction. The thickness of each slice is preferably 4mm, so that the scanning volume extends from the medulla oblongata and the whole brain to the C1–C7 segments of the cervical spinal cord.

[0045] The preferred spatial resolution of the task-state functional images in this invention is a planar resolution of 1.5 mm × 1.5 mm, a repetition time (TR) of 2.6–2.8 s, an echo time (TE) of 25–30 ms, a flip angle of 70–90°, and a foot-to-head axial anteroposterior (e.g., PA) configuration for phase encoding. This is combined with parallel imaging and multi-band excitation (e.g., GRAPPA acceleration factor of 2) to cover the entire brain and cervical spinal cord axis while maintaining temporal resolution. This parameter combination enables the invention to continuously record changes in the BOLD signal induced by standardized thermal pain stimulation in a single scan, ensuring stable and synchronous acquisition of the neural encoding of experimental pain intensity.

[0046] To improve image quality in the brainstem and cervical spinal cord regions, at least one additional set of B0 field maps with the same geometric parameters but opposite phase encoding direction as the functional images are acquired before or during functional image scanning for subsequent EPI distortion correction. Simultaneously, the subjects' respiratory and heart rate signals are synchronously recorded using the physiological monitoring unit of the MRI system at a sampling rate of at least several hundred hertz (e.g., approximately 400 Hz). This provides raw data for subsequent physiological noise regression based on Fourier series modeling, thereby significantly reducing periodic artifacts introduced by cardiac and respiratory movements in the brainstem and cervical spinal cord regions.

[0047] Structural image acquisition preferably employs a three-dimensional T1-weighted sequence (e.g., MPRAGE), with an isotropic resolution of approximately 1 mm³, a TR / TI of approximately 2300 / 900 ms, and a TE of approximately 2–3 ms, to obtain high-quality cortical and cervical spinal cord anatomical information for subsequent brain-spinal cord registration and standardization. These structural and functional image acquisition parameters collectively ensure high spatiotemporal resolution and low-noise synchronous imaging of the entire brain-cervical spinal cord neural activity under experimental pain stimulation tasks, providing a reliable data foundation for the construction of the pain intensity prediction model of this invention.

[0048]

S02

[0049]

S03

[0050] The method for preprocessing brain and cervical spinal cord synchronized task-state functional images is as follows: The scan time was synchronized with respiratory and heart rate signals to create a trigger acquisition signal file and to model physiological noise. This modeling assumed that the physiological processes were quasi-periodic, and respiratory and heart rate data were modeled using low-order Fourier series, generating a total of 32 noise regressors. Global physiological noise was then removed from the corticospinal functional data. Subsequently, distortion correction in the X and Y dimensions of the fMRI images was performed using two corticospinal B0 field images encoded with opposite phases.

[0051]

S04

S02

S03

[0052] The obtained brain and cervical spinal cord data were preprocessed separately, and their structural images were registered to standard templates. The specific method was as follows: For the 3D structural data of the brain, the skull is first dissected, and then the white matter, gray matter and cerebrospinal fluid are segmented.

[0053] For the 3D structural data of the spinal cord, the spinal cord was first automatically segmented to obtain a binary mask, 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 through affine transformation, and then a non-rigid transformation was performed to register the images to the standard spinal cord template (PAM50 T1 weighted template).

[0054] The specific method for registering the brain's task-oriented functional images to structural images is as follows: For task-oriented fMRI images of the brain, head movement correction was first performed. Then, binary masks of white matter and cerebrospinal fluid were registered back to individual functional space and time series were extracted to eliminate the signal influence of white matter and cerebrospinal fluid in the functional image data. The images were further smoothed using a 5 mm full-width half-maximum (FWHM) Gaussian kernel to enhance the signal-to-noise ratio. Then, a high-pass filter with a 100-second truncation period was applied to remove low-frequency signals. Finally, the individual functional space data were aligned to the individual structural space and then rigidly, affinely, and nonlinearly registered to a standard MNI (2 mm) brain template.

[0055] The specific method for registering spinal cord task-oriented functional images to structural images is as follows: For task-oriented fMRI images of the spinal cord, motion correction was first performed, and a cylindrical mask was drawn along the spinal cord midline to exclude extracorporeal regions. Next, spatial smoothing was performed along the x and y axes using 2D Gaussian kernels with an FWHM of 3 mm. To account for noise in the cerebrospinal fluid, the top 10% of variance in the voxel time series was calculated, and regression was performed at each time point using the x and y translations generated by motion correction. Finally, the spinal cord data were registered from individual functional space to individual structural space, and then to the PAM50 standard space.

[0056]

S05

S04

[0057] First-order statistical analysis was performed on the brain and spinal cord task fMRI data registered to standard space. Using a generalized linear model with the pain stimulus paradigm as the main explanatory variable, a contrastive parametric map was generated for each subject. This map quantifies the activation intensity of brain and spinal cord regions in response to pain stimuli. The population average level of the brain and spinal cord activation maps obtained in this step is as follows: Figure 2 As shown in Figure c, it visually illustrates the brain and spinal cord regions (such as the dorsal horn of the C6 segment) that are specifically activated under pain stimulation, which are the signal sources for constructing the predictive model.

[0058]

S06

S061

[0059]

S062

[0060]

S063

[0061] To assess whether the model's predictive power exceeds random probability, a one-sided permutation test was used to evaluate the model's significance. The permutation test was performed by randomly shuffling the participants' pain intensity labels for 10,000 iterations.

[0062] [S064] Perform a generalization test on an independent test dataset to evaluate the significance of the pain intensity prediction model. The final, stable weight space distribution of the pain intensity prediction model is as follows: Figure 3 As shown, it illustrates key regions in the brain and spinal cord that consistently contribute to predicting pain intensity. Figure 3 It is evident that a distributed network involving key nodes such as the anterior / posterior insula (aINS / pINS), dorsal anterior cingulate cortex / supplementary motor area (dACC / SMA), and thalamus (THAL) exhibits a positive correlation between pain intensity and its activity (indicated by warm colors), forming the core positive weighting system driving the model's pain prediction. Simultaneously, the default mode network region, represented by the precuneus, shows negative weights (indicated by cool colors), suggesting that reduced activity in these areas may be related to enhanced pain experience. This weighting map confirms that the biosignals captured by the model have a clear neuroanatomical basis. Example

[0063] Study 1 randomly recruited 92 healthy participants to participate in a simultaneous task-based magnetic resonance imaging experiment of the brain and cervical spinal cord. Study 2 recruited 39 healthy participants as the first independent validation set; Study 3 recruited 58 healthy participants for pain specificity validation.

[0064] All participants were right-handed, in good health, with no history of chronic pain or mental illness, and had not recently taken any analgesics. During the experiment, participants received calibrated thermal pain stimuli applied to their forearms while undergoing a combined brain and cervical spinal cord scan, and reported their instantaneous pain intensity (0-10 numerical rating scale) after each stimulus.

[0065] After preprocessing the experimental data collected simultaneously from the brain and cervical spinal cord, subsequent statistical analysis was performed to construct a pain intensity prediction model (CsPIP model).

[0066] First-order statistical analysis was performed on the brain and spinal cord task-state fMRI data registered to standard space to generate a comparative parameter map of each subject under pain stimulation.

[0067] The obtained brain and spinal cord contrast parameter maps were used as features to build a machine learning pipeline for the development of a pain intensity prediction model.

[0068] To evaluate the generalization ability of the CsPIP model on new data, external validation was performed using the CsPIP model on data from 39 participants in Study 2. The results showed that the CsPIP model effectively predicted pain intensity in Study 2, with a Pearson correlation coefficient of 0.52 (p<0.001). In the pain empathy task of Study 3, the predicted values ​​of the CsPIP model were not significantly correlated with empathy scores, demonstrating its specificity for pain.

[0069] To evaluate the sensitivity of the CsPIP model in clinical intervention, a subset of data receiving transcutaneous electrical nerve stimulation (TENS) was tested using the CsPIP model, such as... Figure 4 As shown in the figure. The results show that the CsPIP model can effectively track the decrease in pain intensity after TENS intervention, and this change only occurs on the treated side.

[0070] Therefore, the pain intensity prediction model based on brain and cervical spinal cord task-state functional connectivity data constructed in this invention can effectively and objectively predict an individual's instantaneous pain intensity, and is suitable for basic research and clinical applications.

[0071] Any aspects not covered in this invention are suitable for use in the prior art. Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A method for constructing a pain intensity prediction model based on task-oriented simultaneous functional magnetic resonance imaging (fMRI), characterized in that, Includes the following steps: Step 1: Induce each subject to a pain stimulus task, collect brain and cervical spinal cord synchronous task-state functional imaging data on the magnetic resonance platform, and convert them into NIFTI format; Step 2: Segment the converted brain and cervical spinal cord synchronous structural images to obtain separate structural image data of the brain and cervical spinal cord; Step 3: After preprocessing the synchronous task-mode functional images of the brain and cervical spinal cord, segmentation is performed to obtain separate task-mode functional image data of the brain and cervical spinal cord. Step 4: Perform subsequent preprocessing on the individual brain and cervical spinal cord data obtained in Step 2 and Step 3, and register their structural images to standard templates and their task-state functional images to structural images. Step 5: Perform first-order statistical analysis on the task-state functional images obtained in Step 4 to generate a comparative parameter map for each subject under pain stimulus conditions. Step 6: Use the comparison parameter map as the input features of the model to build a machine learning pipeline, and develop and validate the prediction model to obtain the pain intensity prediction model.

2. The method for constructing a pain intensity prediction model according to claim 1, characterized in that, The specific method for using the contrast parameter map as the model input feature to construct the machine learning pipeline and develop the pain intensity prediction model in step 6 is as follows: Step 6.1: Standardize the comparison parameter graph; Step 6.2: Use the LASSO-PCR regression model to predict pain intensity; Step 6.3: Use cross-validation to optimize the hyperparameters of the machine learning pipeline in the training set, determine the optimal hyperparameters, and obtain the pain intensity prediction model; Step 6.4: Perform a generalization test on the pain intensity prediction model on an independent test dataset to evaluate the significance of the pain intensity prediction model.

3. The method for constructing a pain intensity prediction model according to claim 1, characterized in that, The brain-cervical spinal cord synchronous magnetic resonance images of each subject were scanned using a 3.0T magnetic resonance imaging system, equipped with a head-neck combined multichannel coil that simultaneously covers the brain, brainstem and cervical spinal cord.

4. The method for constructing a pain intensity prediction model according to claim 1, characterized in that, The task-oriented functional imaging employs a multi-echo planar imaging sequence that simultaneously covers the brain and cervical spinal cord, with a field of view of 192mm × 192mm. Seventy axial slices are arranged along the head-to-tail direction, with a single slice thickness of 4mm, so that the scanning volume extends from the medulla oblongata and the whole brain to the C1–C7 segments of the cervical spinal cord.

5. The method for constructing a pain intensity prediction model according to claim 4, characterized in that, The spatial resolution of the task-state functional images is 1.5mm×1.5mm, the repetition time (TR) is 2.6–2.8s, the echo time (TE) is 25–30ms, the flip angle is 70–90°, the phase encoding direction adopts a foot-head axis anteroposterior configuration, and it combines parallel imaging and multi-band excitation to cover the entire axis of the brain and cervical spinal cord.

6. The method for constructing a pain intensity prediction model according to claim 5, characterized in that, Before or during the task-oriented functional imaging scan, at least one set of B0 field images with the same geometric parameters as the functional images but opposite phase encoding directions are acquired, and the subject's breathing and heartbeat signals are recorded simultaneously.

7. The method for constructing a pain intensity prediction model according to any one of claims 1-6, characterized in that, Synchronous structural imaging of the brain and cervical spinal cord was acquired using a three-dimensional T1-weighted sequence with an isotropic resolution of 1 mm³, a TR / TI of 2300 / 900 ms, and a TE of 2–3 ms.

8. The method for constructing a pain intensity prediction model according to claim 7, characterized in that, In step 2, the coordinates of the upper end of the brain and spinal cord synchronized structural image C1 are first recorded, and then the brain and spinal cord synchronized structural image is divided into two parts: the brain and the spinal cord.

9. The method for constructing a pain intensity prediction model according to claim 7, characterized in that, In step 3, the method for preprocessing the brain and cervical spinal cord synchronized task-state functional images is as follows: Synchronize the scan time with respiratory and heartbeat signals to create a trigger acquisition signal file; The respiratory and heart rate data were modeled using low-order Fourier series to generate several noise regressions, and global physiological noise was removed from the cortical spinal cord function data. Distortion correction of the X and Y dimensions of fMRI images was performed using two cortical spinal cord B0 field images encoded with opposite phases.

10. The method for constructing a pain intensity prediction model according to claim 7, characterized in that, In step 4, the obtained individual brain and cervical spinal cord data are preprocessed separately, and their structural images are registered to a standard template. The specific method is as follows: For the 3D structural data of the brain, the skull is first dissected, and then the white matter, gray matter and cerebrospinal fluid are segmented. For the 3D structural data of the spinal cord, the spinal cord is first automatically segmented to obtain a binary mask of the spinal cord, and then manually corrected according to the segmentation quality. Manually mark the intervertebral discs to straighten the spinal cord; The individual structural image is aligned to the standard vertebral stage through affine transformation; Non-rigid transformations are used to register structural images to standard spinal cord templates.

11. The method for constructing a pain intensity prediction model according to claim 10, characterized in that, In step 4, the task-oriented functional image of the brain is registered to the structural image. The specific method is as follows: Head movement correction was performed on the obtained task-oriented functional images of the brain. The binary mask of white matter and cerebrospinal fluid was registered back to the individual functional space and the time series was extracted; The functional image was smoothed using a 5mm full-width half-maximum FWHM Gaussian kernel. Low-frequency signals are removed by applying a high-pass filter with a 100-second truncation period; Individual functional space data are aligned to individual structural space and rigidly, affinely, and nonlinearly registered to a standard MNI (2 mm) brain template.

12. The method for constructing a pain intensity prediction model according to claim 10, characterized in that, In step 4, the specific method for registering the spinal cord task-oriented functional image to the structural image is as follows: Motion correction was employed, and a cylindrical mask was drawn along the spinal cord axis to exclude areas outside the spinal cord. Spatial smoothing was performed along the x and y axes using a 2D Gaussian kernel with an FWHM of 3 mm. Calculate the top 10% of variances in the voxel time series, and perform regression at each time point using the x and y shifts generated by motion correction; The spinal cord data was registered from the individual functional space to the individual structural space, and then registered to the PAM50 standard space.

Images