Validation of an artificial intelligence-based drawing platform against the Clock Drawing Test for cognitive screening: a comparative study between Alzheimer's disease and healthy controls.

Abstract

Traditional manual Clock Drawing Tests (CDTs) are widely used for cognitive screening but are constrained by human subjectivity and alphanumeric literacy or cultural barriers. This study validated an open-source artificial intelligence (AI)-based drawing platform utilizing universal, pre-linguistic Lea symbols against manual CDT metrics for cognitive screening and differentiating Alzheimer's disease (AD) from healthy controls (HCs). In this cross-sectional validation study, 111 participants aged ≥65 years, 55 clinically diagnosed AD patients and 56 HCs, were evaluated. Participants completed manual CDTs (6-point and 10-point scoring) and the AI-based drawing task on a tablet. Diagnostic accuracy was evaluated using receiver operating characteristic curve analysis, Cohen's kappa (κ), and multivariable logistic regression adjusted for age and gender. The AI model demonstrated clear diagnostic accuracy (area under the curve = 0.868), with overall accuracy comparable to the 10-point manual scale. An optimal diagnostic cut-off of <27.00 points, identified in this study, achieved 83.6% sensitivity and 75.0% specificity, demonstrating substantial agreement with clinical diagnosis (κ = 0.586, P < 0.001). Crucially, the AI-based drawing platform flagged 14 HCs (25.0%) as at-risk who were classified as intact by conventional manual CDT scales. Multivariable regression confirmed the AI score as a powerful independent predictor of AD status after adjusting for demographics (odds ratio: 0.940, 95% confidence interval: 0.904-0.979, P < 0.001). The AI-based drawing platform provides an objective, automated alternative to traditional CDTs. Its sensitivity to subtle visuoconstructional variations positions it as an effective, low-cost triage tool to optimize early neurological referrals in primary care.