Advanced Brain Ageing Associated with Migraine and Post-Traumatic Headache: A Deep Learning Study

Deep learning techniques can estimate "brain age" from MRI as a proxy for brain integrity. The difference between a person's brain and chronological age (Δage) may reflect age-related structural changes. However, brain ageing in headache disorders remains poorly understood. We investigated associations between Δage and headache phenotypes and its potential for predicting headache persistence.

We trained a predictive model of normal ageing using a novel ORdinal Distance Encoded Regularization (ORDER) loss and T1-weighted brain MRI scans from public datasets of healthy controls (HCs; age: 8-95 years). The model outperformed other brain-age methods on a held-out HC test set in our prior work. We then applied this model to an institutional dataset including HCs, participants with acute post-traumatic headache (APTH), persistent post-traumatic headache (PPTH) attributed to mild traumatic brain injury and migraine. The primary outcome was baseline MRI-derived Δage, and its association with headache persistence at follow-up was evaluated.

The study included 137 HC (39.5±12.0 years; 53% female), 99 APTH (42.2±15.3 years; 56.6% female), 49 PPTH (38.1±10.6 years; 34.7% female) and 93 migraine participants (39.6±11.7 years; 74.2% female). Baseline imaging revealed significant differences in predicted Δage across headache phenotypes (F=3.53, ${\eta }_p^2$ =0.03, p<0.05). The HC group showed the smallest predicted Δage (0.46±0.87 years), followed by the APTH (1.25±0.87 years), migraine (3.74±1.03 years), and PPTH (4.65±1.41 years). Among APTH participants, those with headache recovery at 3-months post mTBI had Δage of 1.28±0.22 years compared to 1.63±0.73 years in those without recovery. Participants with headache persistence at 6-month follow-up had Δage of 3.46±0.75 years.

Headache disorders are associated with accelerated brain ageing. Patients with primary (migraine) and secondary (PPTH) headaches exhibit accelerated brain aging and structural decline related to headache persistence compared to healthy controls. APTH participants who later developed headache persistence at 3 and 6 months had older brain signatures at baseline compared to those who recovered. Results demonstrate the potential of Δage as an early neuroimaging marker for identifying individuals at risk for PPTH. Limitations include sample size and lack of longitudinal imaging in all groups.