Self-Supervised Learning with Touchscreen Typing. A Generalizable Strategy for Parkinson's Disease Detection Across Datasets

Shikha Tripathi, Alejandro Acien, Ashley A Holmes, Teresa Arroyo-Gallego, Luca Giancardo

Abstract: Passive monitoring of touchscreen interactions generates keystroke dynamic signals that can be used to detect and track neurological conditions such as Parkinson's disease and psychomotor impairment with minimal burden on the user. However, this is typically done with supervised machine learning methods, requiring datasets that include clinically confirmed labels, which is a challenging task, especially for a large subject pool. In this work, we propose a self-supervised learning (SSL) approach to leverage unlabeled data from control subjects in addition to smaller labeled datasets, typically used when training these types of algorithms.