Carle Illinois Machine Learning System for EEG Analysis Wins IEEE Honors
A new machine learning system developed by a Carle Illinois College of Medicine student could unlock the vast amounts of untapped data found in a common neurological test.
Carle Illinois student Sam Rawal (Class of 2024) collaborated with Yogatheesan Varatharajah, bioengineering research assistant professor, to engineer a solution called SCORE-IT. The system takes in an electroencephalogram (EEG) report without any structure and automatically extracts and standardizes the information.
The team recently won ‘best paper’ honors at the 2021 IEEE Signal Processing in Medicine and Biology Symposium for their publication describing the new system to analyze and classify data from patient EEG tests for use by both clinicians treating patients and researchers seeking out new discoveries.
“The key innovation of our system is that it consists of a multi-stage pipeline that utilizes deep learning for ‘broad parsing’ of the test results, and hand-crafted rules for ‘narrow parsing’ to generate the final classification,” said Rawal.
EEGs are used to diagnose problems like epilepsy by creating a record of a patient’s brainwaves over time and pinpointing abnormalities in the brain’s electrical activity. However, raw waveform records can be extremely large and usually lack standardized labels, making it difficult to detect common patterns found across multiple test results.
“On the clinical side, this standardization makes it easier for clinicians to search and filter patient data based on specific clinical criteria,” Rawal said. The current version of SCORE-IT classifies EEG records based on three criteria: whether the patient is being evaluated for epilepsy, whether the record is normal or abnormal, and the type of seizure detected.
The IEEE Signal Processing award recognizes the engineering work on the SCORE-IT system, but Rawal says his clinical training as a medical student played a key role in developing a successful approach to the lack of standardization in EEG test results. “This problem is prevalent in the United States, and solving it has the potential to improve access to health care data for medical professionals and researchers alike,” Rawal said.
Rawal conducted the research for the paper “SCORE-IT: A Machine Learning Framework for Automatic Standardization of EEG Reports,” in the summer of 2021, as part of Carle Illinois’ Discovery Learning course, in which medical students are immersed in rich research, clinical, or global hands-on learning experiences.