Background & Aims

Does pain modify aging? Machine learning (ML) offers powerful avenues to identify correlates of biological wellbeing that can be influenced by chronic pain. Trigeminal neuralgia (TN)1, the most common form of chronic neuropathic facial pain, serves as a unique and valuable model to better understand the possible dynamic plasticity associated with chronic pain, as it is highly amenable to surgery. Previously, abnormalities in the hippocampus and insula of patients with TN normalized following pain relief from surgery2,3. Patients with TN were also found to have significantly greater brain age than healthy counterparts, with the hippocampal regions contributing significantly to this estimation4. Given the known impact of chronic pain on cognitive and affective functions – areas intrinsically linked to both the hippocampus and the aging process – we hypothesize that hippocampal subfields may serve as more effective predictors of brain age compared with current ML models of brain age.

Methods

522 healthy subjects from Cam-CAN were used to train multiple support vector regression (SVR) models to predict chronological age as a metric defined as brain age. FreeSurfer 7.0 was used to segment T1-weighted magnetic resonance imaging (MRI) scans of whole brains into constituent volumes and hippocampal subfield volumes. Separate SVRs were trained on only the hippocampal subfields, whole-brain volumes, and a combination of both, to compare the accuracy of their predictions. Subjects were split into 80-20% training and testing cohorts before fitting each model to their respective training cohorts. 10-fold cross-validation was performed to calculate the average prediction accuracy of each model’s brain age outputs before they were run on the testing cohorts. Models were then run on 123 TN patients. Performance metrics, such as the correlation coefficient (r), coefficient of determination (R2), mean absolute error (MAE), and root-mean-square error (RMSE) are reported.

Results

First, an SVR built on raw hippocampal subfield volumes predicted brain ages with a strong, positive correlation with chronological ages (r = 0.83, p < 0.001; R2 = 0.67; MAE = 8.78; RMSE = 11.22). After adjusting for varying head size, a separate SVR was trained on hippocampal subfields, which performed slightly better (r = 0.88, p < 0.001; R2 = 0.76; MAE = 7.71; RMSE = 9.60) and nearly as well as one built on whole-brain segmentations (r = 0.91, p < 0.001; R2 = 0.82; MAE = 6.64; RMSE = 8.35). Lastly, an SVR built on a combination of whole-brain volumes and hippocampal subfields outperformed all previous models (r = 0.92, p < 0.001; R2 = 0.84; MAE = 6.36; RMSE = 7.81). The final combined model was used to predict the ages of individuals with TN (r = 0.77, p < 0.001; R2 = 0.55; MAE = 7.81; RMSE = 9.77) and found their mean brain age, 61.65 ± 0.84 years (standard error of the mean), to be significantly greater than their actual mean age of 59.59 ± 1.32 years (t122 = -2.38, p = 0.019).

Conclusions

Advances in ML have ushered in the conceptualization of brain age as a robust predictor of various other clinically-relevant variables. Generally, across neuropsychiatric, metabolic, developmental, and neurodegenerative diseases, a brain age greater than chronological age correlates with risk of developing chronic illnesses, cognitive decline, and mortality8,8–11. Accordingly, we present another instance in which patients with TN, a debilitating manifestation of chronic neuropathic facial pain, appear to have brains that are biologically older than their actual age. The open-access methods to compute brain age from MRI data used in this study supports adoption of brain age as a clinically-relevant biomarker that may assist clinicians in deciding therapeutic routes for their patients. A future direction would be to calculate the brain age of TN patients following pain-relieving surgical intervention to reveal whether accelerated biological aging can be prevented or possibly reversed.

References

1. Cruccu, G., Di Stefano, G. & Truini, A. Trigeminal Neuralgia. N. Engl. J. Med. 383, 754–762 (2020).
2. Noorani, A. et al. Pain Relief Reverses Hippocampal Abnormalities in Trigeminal Neuralgia. J. Pain 23, 141–155 (2022).
3. DeSouza, D. D., Davis, K. D. & Hodaie, M. Reversal of insular and microstructural nerve abnormalities following effective surgical treatment for trigeminal neuralgia. Pain 156, 1112–1123 (2015).
4. Hung, P. S.-P. et al. Differential expression of a brain aging biomarker across discrete chronic pain disorders. PAIN 163, 1468 (2022).
5. Zhang, X. et al. Evidence for Cognitive Decline in Chronic Pain: A Systematic Review and Meta-Analysis. Front. Neurosci. 15, 737874 (2021).
6. Fine, P. G. Long-Term Consequences of Chronic Pain: Mounting Evidence for Pain as a Neurological Disease and Parallels with Other Chronic Disease States. Pain Med. 12, 996–1004 (2011).
7. Vadivelu, N. et al. Pain and Psychology—A Reciprocal Relationship. Ochsner J. 17, 173–180 (2017).
8. Angoff, R. et al. Relations of Metabolic Health and Obesity to Brain Aging in Young to Middle?Aged Adults. J. Am. Heart Assoc. Cardiovasc. Cerebrovasc. Dis. 11, e022107 (2022).
9. Cole, J. H., Marioni, R. E., Harris, S. E. & Deary, I. J. Brain age and other bodily ‘ages’: implications for neuropsychiatry. Mol. Psychiatry 24, 266–281 (2019).
10. Drobinin, V. et al. The Developmental Brain Age Is Associated With Adversity, Depression, and Functional Outcomes Among Adolescents. Biol. Psychiatry Cogn. Neurosci. Neuroimaging 7, 406–414 (2022).
11. Mattson, M. P. & Arumugam, T. V. Hallmarks of Brain Aging: Adaptive and Pathological Modification by Metabolic States. Cell Metab. 27, 1176–1199 (2018).
12. Fischl, B. FreeSurfer. NeuroImage 62, 774–781 (2012).
13. Iglesias, J. E. et al. A computational atlas of the hippocampal formation using ex vivo, ultra-high resolution MRI: Application to adaptive segmentation of in vivo MRI. NeuroImage 115, 117–137 (2015).
14. Iglesias, J. E. et al. Bayesian longitudinal segmentation of hippocampal substructures in brain MRI using subject-specific atlases. NeuroImage 141, 542–555 (2016).
15. Saygin, Z. M. et al. High-resolution magnetic resonance imaging reveals nuclei of the human amygdala: manual segmentation to automatic atlas. NeuroImage 155, 370–382 (2017).
16. Hung, P. S.-P. et al. Sex differences in patient journeys to diagnosis, referral, and surgical treatment of trigeminal neuralgia: implications for equitable care. J. Neurosurg. 1–9 (2022).

Presenting Author

Jerry Li

Poster Authors

Jerry Li

BSc(Hons)

Institute of Medical Science, University of Toronto

Lead Author

Timur Latypov

University of Toronto

Lead Author

Patcharaporn Srisaikaew (PhD)

Krembil Brain Institute, University Health Network

Lead Author

Mojgan Hodaie

University of Toronto and Toronto Western Hospital

Lead Author

Topics

  • Pain Imaging