Fig. 1: High-level view of integrating deep learning training within the clinical workflow.
From: Weak supervision as an efficient approach for automated seizure detection in electroencephalography
After EEG monitoring on patients, a mixed group annotates the signal before a clinician does the final assessment to generate a medical report. In our work, we evaluate using weak labels, which are fast and cheap to acquire within the clinical workflow, to train deep learning models for onset seizure detection. On the other hand, existing methods use expert-provided gold labels, which are costly and not a part of the clinical workflow.
