An investigation of the contextual distribution of false positives in a deep learning-based atrial fibrillation detection algorithm.