Techniques for visualizing LSTMs applied to electrocardiograms
This work addresses interpretability for medical AI applications, but it is incremental as it compares existing visualization methods on a specific domain.
This paper tackled the problem of visualizing LSTM networks for time series data, specifically electrocardiograms, and found that an input deletion mask technique best reduced class scores, with features aligning with medical theory on the MIT-BIH dataset.
This paper explores four different visualization techniques for long short-term memory (LSTM) networks applied to continuous-valued time series. On the datasets analysed, we find that the best visualization technique is to learn an input deletion mask that optimally reduces the true class score. With a specific focus on single-lead electrocardiograms from the MIT-BIH arrhythmia dataset, we show that salient input features for the LSTM classifier align well with medical theory.