How can the latest advances in data science and AI help people with epilepsy?
We think there are many ways!
Building advanced simulation tools
Simulations are developed at multiple levels of abstraction, including: small collections of neurons to understand how short term memory works, large scale MEG and EEG simulations to help understand the limitations of each modality, realistic seizure diary simulations, and clinical trial simulations.
What machine learning (ML) tools can do for epilepsy trials?
Can ML analyze a clinical trial more efficiently than traditional methods? (maybe yes!) Can machine learning help predict the outcome of a clinical trial even before it is done (maybe yes!)
Can AI be used to automate EEG interpretation?
With our collaborator Dr. Brandon Westover and his team, we think yes!
How can large language models (LLMs) be used in clinical care?
We showed that multiple LLMs can pass 8 practice Epilepsy Board exams! We also simulated a case where LLMs are used to show inductive reasoning based on clinical notes.
Re-examining basic questions about epilepsy
Can a patient swallow their tongue during a seizure? (no!) Does sleep deprivation cause seizures? (maybe not!) Do only women have seizure risk cycles? (men do too!) What is a seizure cluster? (it is probably patient-specific, but we can calculate it!) Did that drug improve that patient? (it can be calculated more accurately than doctor's guesses) How long do typical self-reported seizures last? (longer than the textbook says!)
Using statistics to understand seizure patterns
Do seizures have risk patterns on different timescales? (yes!) What models should we use to generate synthetic seizure diaries? (overdispersed ones) How often do patients have seizure clusters? (depends mainly on your definition!) What are usual seizure rates? (we examined >10,000 patient diaries to answer this). Is there a relationship between average seizure rate and variability in seizure rate? (yes! we call it the L relationship).
Using data science to think about biosensors
How accurate does a seizure detector need to be in order to be useful? Can a pulse oximeter be used in the hospital to detect seizures? How efficient would a clinical trial be if patients had electrodes implanted near their seizure focus?
