TY - JOUR
T1 - Evaluating Deep Learning Performance for P300 Neural Signal Classification
AU - Ravipati, Yashwanth
AU - Pouratian, Nader
AU - Arnold, Corey
AU - Speier, William
N1 - Publisher Copyright:
©2023 AMIA - All rights reserved.
PY - 2023
Y1 - 2023
N2 - P300 event-related potential (ERP) signals are useful neurological biomarkers, and their accurate classification is important when studying the cognitive functions in patients with neurological disorders. While many studies have proposed models for classifying these signals, results have been inconsistent. As a result, a consensus has not yet been reached on the optimal model for this classification. In this study, we evaluated the performance of classic machine learning and novel deep learning methods for P300 signal classification in both within and across subject training scenarios across a dataset of 75 subjects. Although the deep learning models attained high attended event classification F1 scores, they did not outperform Stepwise Linear Discriminant Analysis (SWLDA) in the within-subject paradigm. In the across-subject paradigm, however, EEG-Inception was able to significantly outperform SWLDA. These results suggest that deep learning models may provide a general model that do not require subject-specific training and calibration in clinical settings.
AB - P300 event-related potential (ERP) signals are useful neurological biomarkers, and their accurate classification is important when studying the cognitive functions in patients with neurological disorders. While many studies have proposed models for classifying these signals, results have been inconsistent. As a result, a consensus has not yet been reached on the optimal model for this classification. In this study, we evaluated the performance of classic machine learning and novel deep learning methods for P300 signal classification in both within and across subject training scenarios across a dataset of 75 subjects. Although the deep learning models attained high attended event classification F1 scores, they did not outperform Stepwise Linear Discriminant Analysis (SWLDA) in the within-subject paradigm. In the across-subject paradigm, however, EEG-Inception was able to significantly outperform SWLDA. These results suggest that deep learning models may provide a general model that do not require subject-specific training and calibration in clinical settings.
UR - http://www.scopus.com/inward/record.url?scp=85182547860&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85182547860&partnerID=8YFLogxK
M3 - Article
C2 - 38222383
AN - SCOPUS:85182547860
SN - 1559-4076
VL - 2023
SP - 1218
EP - 1225
JO - AMIA ... Annual Symposium proceedings. AMIA Symposium
JF - AMIA ... Annual Symposium proceedings. AMIA Symposium
ER -