Abstract

Migraines are a complicated neurological illness that affects millions worldwide, its diagnosis is problematic due to dependence on subjective symptomatology and the absence of reliable biomarkers. This study aims to utilize Artificial Intelligence (AI) to improve migraine diagnosis based on tabular clinical data. Two scenarios were adopted: in the first scenario, a single-stage, multi-class Deep Neural Network (DNN) model was used along with several ML algorithms to classify migraine into four classes. In the second scenario, a two-stage hierarchical framework based on DNN was developed. The first stage (S1) is binary classification to classify (Migraine vs non-migraine, while the second stage (S2) classifies the migraine cases into their three common subtypes, providing an innovative AI-based diagnostic tool that mimics clinical diagnostic processes. The results revealed that the DNN model in the first scenario achieved the best performance with an accuracy of 99.55%. In the second scenario, S1 and S2 tests achieved high performance with accuracy ≈ 99.55, 93.75 respectively. In order to assess generalizability, the hierarchical model was evaluated using an external dataset. The generalization accuracy of S1 reached 96.5% while S2 reached 96.61%. In summary, although the first case outperforms well in terms of the classification performance using the source dataset, our second proposed hierarchical framework not only achieves high classification accuracy but also have a strong ability of generalization for more realistic clinical applications. These findings may lead to a more accurate, faster, and cost-effective diagnosis of migraine in clinical practice.