Explainable Deep Learning Approach For Extracting Cognitive Features From Hand-drawn Images Of Intersecting Pentagons

This article discusses the use of deep learning to predict global cognition performance from intersecting pentagon drawings. Data from 3111 participants from three ongoing cohort studies of aging and dementia at the Rush Alzheimer’s Disease Center (RADC) was used to train a deep learning model. 47 established deep-learning models for vision recognition were evaluated and an architecture that demonstrated high and robust performance was identified. A pentagon-drawing simulator was developed to interrogate the deep learning model and suggest key drawing characteristics in people with lower cognitive function.