Let's Hope it Works! Inaccurate Supervision of Neural Networks with Incorrect Labels: Application to Epilepsy

Florian Dubost, Erin Hong, Daniel Y Fu, Nandita Bhaskhar, Siyi Tang, Khaled Saab, Daniel Rubin, Jared Dunnmon, Christopher Lee-Messer

This work describes multiple weak supervision strategies for video processing with neural networks in the context of epilepsy. To study seizure onset, researchers have designed automated methods to detect seizures from electroencephalography (EEG), a modality used for recording electrical brain activity. However, the EEG signal alone is sometimes not enough for existing detection methods to discriminate seizure from artifacts having a similar signal on EEG. For example, such artifacts could be triggered by patting, rocking or suctioning in the case of neonates. In this article, we addressed this problem by automatically detecting an example artifact-patting of neonates -- from continuous video recordings of neonates acquired during clinical routine. We computed frame-to-frame cross-correlation matrices to isolate patterns showing repetitive movements indicative of patting. Next, a convolutional neural network was trained to classify whether these matrices contained patting events using weak training labels -- noisy labels generated during normal clinical procedure. The labels were considered weak as they were sometimes incorrect. We investigated whether networks trained with more samples, containing more uncertain and weak labels, could achieve a higher performance. Our results showed that, in the case of patting detection, such networks could achieve a higher recall and focused on areas of the cross-correlation matrices that were more meaningful to the task, without sacrificing precision.

Knowledge Graph



Sign up or login to leave a comment