A Novel Temporal Variability-Aware Architecture for Ictal-Interictal Continuum Patterns Classification
Year: 2026
Type: preprint
Abstract: Harmful brain activity may compromise cerebral function and can culminate in mortality if not addressed appropriately. The global deficit of professionals adept in electroencephalogram (EEG) data analysis exacerbates the difficulty in resolving this issue. There is a pressing need for a reliable and automated system, particularly designed to effectively classify harmful brain activity using both signals and images. In this study, we proposed a novel model that can be used for signal as well as image classification. This study employs EEG recordings obtained from the HMS dataset and MRI scans from the Brain Tumor MRI dataset. The EEG dataset comprises 106,800 unique EEG identification numbers. Our proposed model, EfficientNetZEEGv1, with instability-aware gated TSA, demonstrated exceptional performance by achieving a validation accuracy of 97% and a loss of 0.20. We made some enhancements in the gating mechanism of the model, and our second model EfficientNetZEEGv2, achieved a validation accuracy of 98.74%. We have achieved the highest accuracy, precision, and recall values amongst all studies conducted in the domain of harmful brain activities.
Source: Research Square
Language: English
Subfield: Cognitive Neuroscience
Field: Neuroscience
Domain: Life Sciences
Open Access status: green
OpenAlex ID: W7237217850
