Figure 1.

Study methods.

A) Data used in the study comprising the ‘brain-age’ training sample of n=2,646 healthy individuals and the current experiment cohort (n=47) comprising those with chronic pain (n=33) and those without (n=14).

B) Image pre-processing, applied to all images, used SPM12 software to segment T1-weighted MRIs into gray and white matter probability maps. These were then spatially-normalized using DARTEL non-linear registration to a custom template in MNI152 space, with 1.5mm³ voxels, using 4mm spatial smoothing. These normalized 3D images were converted into 1D vectors and the gray and white matter vectors concatenated.

C) Machine learning age prediction involved generating a linear kernel by calculating the dot-product of all pairs of image vectors across all participants, resulting in a similarity matrix. The similarity matrix was used as input into a Gaussian Processes regression to predict chronological from the image vectors. The model trained on the full training set was then applied to the n=47 participants from the chronic pain study to generate a brain-predicted age value for each participant.

D) Statistical analysis was conducted to evaluate performance of the regression model performance using ten-fold cross-validation. Brain-predicted age difference (brain-PAD) was then calculated for the chronic pain study participants; whereby chronological age was subtracted from brain-predicted age. Brain-PAD was then used for subsequent statistical analysis of pain-related variables.