Despite the nationwide shortage of neurologists, the present-day neurological care relies heavily on time-consuming visual review of patient data by trained neurologists. This is particularly emphasized in the field of epileptology where epileptologists spend a substantial amount of their time on visually reviewing and interpreting lengthy multi-channel time series of brain electrical activity, called electroencephalography (EEG). This burden not only contributes to the escalation of epileptologist burnout, but also introduces reviewer bias and potential errors in clinical decisions. In fact, several studies have reported that the inter-reviewer agreement in reviewing EEGs is surprisingly poor and that the sensitivity of EEG-based expert visual diagnosis of epilepsy remains at 50%. As such, the field of neurology needs solutions that can, a) augment expert visual review of EEGs, b) reduce the workload, and c) enhance the reliability, reproducibility, and scalability of EEG review.

This project focused on developing a machine-learning (ML)-based decision support framework to actionable information extracted from lengthy EEG recordings using domain-guided machine learning. We successfully demonstrated the value of this framework by introducing significant innovations across the data-science lifecycle and by evaluating its utility in diagnosing epilepsy and cognitive diseases. Our technical innovations included, a) development of domain-guided and interpretable ML methods to augment EEG review and analysis; b) development of automated natural language processing tools to parse EEG reports and generate useful metadata; and c) developing novel self-supervised pretraining and curriculum learning approaches for EEG ML model training. We also collaborated with clinicians at the Mayo Clinic and the Mayo Clinic Neurology Artificial Intelligence program to demonstrate the clinical value using real-world clinical data collected from patients with various neurological diseases. Overall, this framework reduces the need for visual review of EEG data and increases the availability of clinicians to patient care.

This award was used to partially support two PhD students during the initial years of their training. We also engaged with UIUC undergraduate students and medical students at the Carle Illinois College of Medicine to expose them to novel ML methods in clinical applications. In addition, we participated in the summer K-12 research program called “Worldwide Youth in Science and Engineering” and trained two high school students on EEG recording using a wireless device and python programming. This training provided students with the necessary experience to obtain college/university placements, full-time jobs, clinical residency, and internships. The outcomes of this project were published in peer-reviewed conference proceedings and medical journals, and software tools were released as github repositories. PI Varatharajah also received an NSF CAREER award focusing on making EEG-ML decision support tools more robust and trustworthy, which directly builds on the work supported by this award.

Last Modified: 01/02/2025
Modified by: Yogatheesan Varatharajah