EXPLORING THE USE OF MACHINE LEARNING TO IMPROVE DEMENTIA DETECTION: AUTOMATING CODING OF THE CLOCK-DRAWING TEST

Abstract Alzheimer’s disease and related dementias, a leading cause of disability among older adults, has become critical public health concern. The clock-drawing test (CDT), which asks subjects to draw clock, typically with hands showing 11:10, been widely used as screening tool detect dementia in clinical research surveys. clock-drawings are often coded into binary (e.g., normal vs. abnormal) or ordinal scores. A limitation large-scale studies is that the manual-coding CDT could result biases if coders interpret implement coding rules differently. Several small-scale have explored use machine learning methods automate coding. Such either had limited success not designed specifically for complex images. This study aims create evaluate an intelligent CDT-scoring system automatically codes scores using deep neural networks (DLNN) including resnet101, EfficientNet-B3, Visual Transformer. We large repository images from 2011-2019 National Health Aging Trends Study, panel Medicare beneficiaries ages 65 older. Results show 1) DLNN achieved high scoring accuracy; 2) generates more consistent accurate than trained lay coders, compared expert neuropsychology background; 3) DLNN-coded highly correlated self-reported diagnosis performance-based assessments memory. offers model automating other drawing tests variety cognitive functions.