Scope and arbitration in machine learning clinical EEG classification

Zhu, Yixuan; Canham, Luke; Western, D.

Abstract

A key task in clinical EEG interpretation is to classify a recording or session as normal or abnormal. In machine learning approaches to this task, recordings are typically divided into shorter windows for practical reasons, and these windows inherit the label of their parent recording. We hypothesised that window labels derived in this manner can be misleading - for example, windows without evident abnormalities can be labelled 'abnormal' - disrupting the learning process and degrading performance. We explored two separable approaches to mitigate this problem: increasing the window length and introducing a second-stage model to arbitrate between the window-specific predictions within a recording. Evaluating these methods on the Temple University Hospital Abnormal EEG Corpus, we significantly improved state-of-the-art average accuracy from 89.8 percent to 93.3 percent. This result defies previous estimates of the upper limit for performance on this dataset and represents a major step towards clinical translation of machine learning approaches to this problem. Our study includes electroencephalography (EEG) datasets collected from https://isip.piconepress.com/projects/tuh\-eeg/. Our code is shared on https://github.con zhuyixuanl997/ EEGScopeAndArbitration.

• Presentation Conference Type: Conference Paper (published)
• Conference Name: The 2023 IEEE Signal Processing in Medicine and Biology Symposium
• Start Date: Dec 1, 2023
• Acceptance Date: Oct 1, 2023
• Online Publication Date: Dec 29, 2023
• Publication Date: Dec 29, 2023
• Deposit Date: Dec 13, 2023
• Publicly Available Date: Dec 30, 2025
• Publisher: Institute of Electrical and Electronics Engineers (IEEE)
• Book Title: 2023 IEEE Signal Processing in Medicine and Biology Symposium (SPMB)
• ISBN: 9798350341263
• DOI: https://doi.org/10.1109/SPMB59478.2023.10372635
• Keywords: electroencephalogram; abnormality detection; deep learning
• Public URL: https://uwe-repository.worktribe.com/output/11515342
• Additional Information: Currently published on conference. Should be indexed and available through IEEE Explore in a few weeks.