Machine learning model using imaging data predicts drug-refractory trigeminal neuralgia with moderate accuracy

SVM-RBF model integrating clinical, pain, and imaging data predicted drug-refractory trigeminal neuralgia with test AUC 0.824 and validation AUC 0.806; Cox models showed moderate short-to-medium discrimination but limited long-term accuracy.

• SVM-RBF achieved the highest average AUC among five machine learning models (training 0.927, test 0.824, validation 0.806).
• Pain involved extent and medial temporal lobe atrophy (MTA) score were independent prognostic markers in Cox regression.
• Long-term predictive performance attenuated, highlighting the need for improved survival modeling in this setting.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.