Background & Aims

Patients with multiple sclerosis (MS) have a twenty-fold greater risk of developing trigeminal neuralgia (TN), a severe form of facial neuropathic pain[1]. In patients who developed TN secondarily to MS (MS-TN), objective neural signatures of pain remain elusive. Previous research suggests that gray matter abnormalities may be crucial indicators of pain, prompting the investigation of their role in the TN pain experienced by MS patients for the first time[2]. However, pain in MS is understudied and may be difficult to investigate particularly in those with advanced disease [3]. This suggests the need for identifying the imaging signatures of pain in MS. In this work we use machine learning (ML) methods to examine differences between the gray matter of MS and MS-TN patients. This analysis would allow us to identify key imaging predictors of TN at the levels of regional cortical thickness and area, and subcortical volume.

Methods

We analyzed T1-weighted MR imaging data from 75 MS and 77 MS-TN patients, matched by age, sex, and MS duration. We processed imaging data using Freesurfer 7 and extracted cortical and subcortical gray matter metrics [4]. Using a support vector machine classifier, we trained a machine learning model to predict presence of TN pain in MS patients using their imaging metrics alone. We used 10-fold nested cross-validation and feature selection to optimize set of imaging predictors and perform out-of-sample evaluation.

Results

Our ML classifier distinguished between MS and MS-TN imaging metrics on average achieving 99% of training and 93.4% testing accuracy with high recall for both classes. Structures within visual, attentional, somatomotor, and default-mode networks (hippocampus, thalamic subnuclei, occipital cortex and postcentral gyrus) were identified as significant imaging predictors of TN pain in MS.

Conclusions

We were able to accurately distinguish between MS and MS-TN using ML analysis, which pointed towards several key imaging predictors of TN pain in MS patients. Our results emphasize the multifaceted nature of pain and pave the way towards using imaging to assess and understand pain disorders with greater objectivity.

References

1.Di Stefano G (2019). Trigeminal neuralgia secondary to multiple sclerosis: from the clinical picture to the treatment options. The journal of headache and pain, 20(1), 20.
2.Latypov TH (2023). Brain imaging signatures of neuropathic facial pain derived by artificial intelligence. Scientific reports, 13(1), 10699.
3.Kurtzke JF. (1983). Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (EDSS). Neurology, 33(11), 1444–1452.
4.Dale AM (1999). Cortical surface-based analysis. I. Segmentation and surface reconstruction. Neuroimage 9, 179-194

Presenting Author

Timur Latypov

Poster Authors

Timur Latypov

MD

University of Toronto

Lead Author

E-mail

Abigail Wolfensohn

McGill University

Lead Author

E-mail

Rose Yakubov

University of Toronto

Lead Author

E-mail

Jerry Li

Institute of Medical Science, University of Toronto

Lead Author

E-mail

Daniel Jörgens

University Health Network

Lead Author

E-mail

Patcharaporn Srisaikaew (PhD)

Krembil Brain Institute, University Health Network

Lead Author

E-mail

Ashley Jones

Unity Health Toronto

Lead Author

E-mail

Errol Colak

Unity Health Toronto

Lead Author

E-mail

David Mikulis

University Health Network

Lead Author

E-mail

Frank Rudzicz

Vector Institute for Artificial Intelligence

Lead Author

E-mail

Jiwon Oh

Unity Health Toronto

Lead Author

E-mail

Mojgan Hodaie

University of Toronto and Toronto Western Hospital

Lead Author

E-mail

Topics

  • Pain Imaging