Machine learning-based migraine analysis using retinal vessel diameters from optical coherence tomography: an alternative approach.

objectiveMigraine is a primary headache disorder characterised by attacks of headache that are usually unilateral and throbbing in nature, may be accompanied by neurological symptoms, and, due to its complex pathophysiology, can affect not only the central nervous system but also structures such as the retinal vascular system. In recent years, retinal imaging techniques have emerged as a promising method for studying neuro-ophthalmological diseases. In this study, we aimed to predict migraine by evaluating the measurements made from retinal images obtained with Optical Coherence Tomography (OCT). MATERIALS AND

methodsIn the present study, 70 eyes of migraine patients and 38 eyes of healthy control group were examined. In cases where there was an imbalance between the classes, the data were balanced by applying the SMOTE method, which is widely preferred in studies. In addition to age and gender data, features such as retinal artery and vein diameters and choroidal thickness measurements were used as data. Pearson's Correlation Coefficient method was applied to calculate the linear relationship between the features.

resultsClassification results were evaluated with Area Under the Curve (AUC), Accuracy (Acc), Kappa statistic (KS), F1-score (F1), and Matthews Correlation Coefficient (MCC) parameters. The most successful result in the classification process between migraine and healthy control was obtained with the LightGBM algorithm with 93.28% AUC, 91.14% Acc, 86.67% F1, 0.74 KS, and 0.76 MCC rates.

conclusionThe presented research can be considered as a preliminary study. The results of the research on the application of machine learning algorithms showed an effective performance in migraine prediction from OCT data. Ensemble-based Boosting model classifiers were more successful than traditional machine learning classifiers.