Machine Learning in Epilepsy Drug Treatment Outcome Prediction Using Multi-modality Data in Children with Tuberous Sclerosis Complex

Epilepsy drug treatment outcome prediction is valuable for the treatment process of children with tuberous sclerosis complex. In this paper, three common feature selection methods and six common machine learning models are used to predict epilepsy drug treatment outcomes with multi-modality data in children with tuberous sclerosis complex. The analysis of variance F-value selecting 35 features combined with multilayer perceptron achieves the best area-under-curve score (95% confidence interval) of 0.812 (± 0.005), which shows the feasibility of using machine learning to predict the outcomes of drug treatments. Then, the effectiveness of the lesion features in the magnetic resonance imaging is evaluated from 3 important perspectives: quantity, location, and type. Our analysis results found that among them, lesion type is the most important in the outcome prediction, followed by location and quantity.