Machine learning models for personalised epilepsy management

The current standard of care for epilepsy is to commence treatment with a single antiseizure medication (ASM) and if it fails, try successive drug regimens until the epilepsy is regarded as ‘pharmacoresistant’. This trial-and-error approach does not consider factors beyond broad seizure types or epilepsy syndromes in drug selection, and risks damage to the brain from uncontrolled seizures while patients endure trials of ineffective treatments, and 30% patients remaining uncontrolled. There is no reliable way to predict which drug will work best for a given patient, nor to predict whether their epilepsy is likely drug-resistant. As a result, non-drug options, such as surgery, are generally reserved as a last resort, by which time they might be less effective, and irreversible damage might have occurred. We hypothesise that it is possible to predict response to ASM treatment and identify individuals with high risk of pharmacoresistant epilepsy using machine learning techniques. We have recently developed a world-first machine learning model to predict the most effective initial ASM using routinely collected clinical data from patients in UK and Perth. In this project we will improve the performance of our model by integrating it with diagnostic reports and genetic information, extend the model to predict response to subsequent drug regimens, and to predict DRE at the initial diagnosis of epilepsy. We will validate these models in a separate cohort of new onset epilepsy patients in Melbourne. The models will allow personalised treatment for patients with epilepsy, leading to improved outcomes. This project will suit students with interest in data science, application of artificial intelligence in medicine, epidemiology, genomics.