Electroencephalography Biomarkers to Characterize and Predict Parkinson’s Disease using Cycle-By-Cycle Analysis of Neural Oscillations and a Deep Learning Model

Started Aug. 30, 2022

Abstract or project description

Parkinson's Disease (PD) is a prevalent and challenging neurodegenerative disorder, affecting nearly 10 million people globally. Current diagnostic methods rely on observations of symptoms and medical history but have proven to be fallible, with misdiagnosis rates ranging from nearly 10%-20%. Furthermore, a lack of understanding in the field of PD poses a great hurdle in determining a well-fitted treatment for each patient. As of now, PD has no cure, making early detection critical for appropriate intervention to slow its progression. Electroencephalograms (EEGs) are a novel approach in the field of neurodegenerative diseases that provide crucial information on a patient's brain’s electrical activity, which can be crystalized into new observable symptoms of PD, as well as serve in diagnosing the disease earlier and more accurately. In this project, we verify the applicability of EEG biomarkers to characterize and predict PD using a cycle-by-cycle analysis of neural oscillations. We segment patient recordings into individual cycles, which offer a more accurate measure than conventional methods that operate on a sinusoidal basis, and compute relevant features. A statistically significant difference (p<0.05) between PD and Control in amplitude, period, and symmetry suggests that EEGs are able to differentiate the two groups successfully. To verify these findings, we implement a sequential neural network model to classify a segmented filtered reading, which attained a 96.15% accuracy, well outperforming current methodologies. These results indicate that electro-neurophysiological data has strong potential in effectively predicting PD earlier, ultimately improving treatment outcomes for millions worldwide.