Learned Kernels for Sparse, Interpretable, and Efficient Medical Time Series Processing

Background: Rapid, reliable, and accurate interpretation of medical signals is crucial for high-stakes clinical decision-making. The advent of deep learning allowed for an explosion of new models that offered unprecedented performance in medical time series processing but at a cost: deep learning models are often compute-intensive and lack interpretability. Methods: We propose Sparse Mixture of Learned Kernels (SMoLK), an interpretable architecture for medical time series processing. The method learns a set of lightweight flexible kernels to construct a single-layer neural network, providing not only interpretability, but also efficiency and robustness. We introduce novel parameter reduction techniques to further reduce the size of our network. We demonstrate the power of our architecture on two important tasks: photoplethysmography (PPG) artifact detection and atrial fibrillation detection from single-lead electrocardiograms (ECGs). Our approach has performance similar to the state-of-the-art deep neural networks with several orders of magnitude fewer parameters, allowing for deep neural network level performance with extremely low-power wearable devices. Results: Our interpretable method achieves greater than 99% of the performance of the state-of-the-art methods on the PPG artifact detection task, and even outperforms the state-of-the-art on a challenging out-of-distribution test set, while using dramatically fewer parameters (2% of the parameters of Segade, and about half of the parameters of Tiny-PPG). On single lead atrial fibrillation detection, our method matches the performance of a 1D-residual convolutional network, at less than 1% the parameter count, while exhibiting considerably better performance in the low-data regime, even when compared to a parameter-matched control deep network.