Machine Learning Approach to Predict Cognitive Domains with Process Parameters in Digital Drawing Tasks

Digital drawing data offers a promising, time-efficient addition to the neuropsychological assessment of neurodegenerative diseases, providing a more detailed understanding of early deficits and helping to differentiate neurodegenerative disease subtypes. This study investigates the predictive power of digitally drawing data for domain-level cognitive deficits. We analyzed a single time point assessments from 937 participants in a longitudinal observational study. We extracted 5 data-driven cognitive domains from a neuropsychological assessment battery using affinity propagation: attention, executive/language, intrusions, memory, and visuospatial, which corresponded with conventional theory-driven domains and differentiate between healthy individuals and those with mild cognitive impairment. Machine learning used to predict these data-driven domains from process parameters of digital drawing data showed substantial predictive value for the executive/language domain and moderate value for the memory domain. These findings underscore the potential of digital drawing data for time-efficient diagnostic purposes, particularly in assessing time-sensitive and graphomotoric facets of cognitive impairment.