Figure 3

Visualization of the important MR imaging classification predictors identified by AI models, arranged according to the classification task and side of pain (ipsilateral versus contralateral to pain side). (A) CTN versus HC models, (B) TNP versus HC models, (C) CTN versus TNP models. Middle column depicts coronal views. Vertices correspond to the cortical thickness and surface area metrics, lines correspond to the diffusivity metrics (FA, RD, AD), and circles correspond to the thalamic nuclei volume metrics. A total of 12 neuroanatomical predictors were visualized for both LR and RF CTN versus HC models, of which 4 were bilateral, 7 contralateral and one ipsilateral. Also, 12 predictors were visualized for both LR and RF TNP versus HC models, of which 3 are bilateral, 7 were contralateral and 2 ipsilateral. Only two predictors are identified for CTN versus TNP, one contralateral and one ipsilateral. All displayed neuroanatomical predictors showed significant differences between compared groups (CTN versus HC, TNP versus HC and CTN versus TNP respectively). HC healthy control, CTN classical trigeminal neuralgia, TNP trigeminal neuropathic pain, HC healthy control. B Bilateral, I Ipsilateral to pain side, C Contralateral to pain side. Rendered in Paraview.