AI-Enhanced Animal Naming as a Digital Biomarker for Early Cognitive Screening

The animal naming task is a widely used, low-burden measure of semantic fluency for cognitive screening, but conventional scoring based on total correct items overlooks linguistic and temporal features relevant to early mild cognitive impairment (MCI). We developed a multi-dimensional, AI-enhanced scoring framework that quantifies base count and age-adjusted percentiles, efficiency (unique/repetition balance and diversity), semantic flexibility (category switching and coverage), and optional speech-quality features when transcripts are derived from audio. In a proof-of-concept simulation, we generated text responses emulating 60-second performances from cognitively healthy and impaired profiles. A supervised classifier integrated the feature families into a composite score with confidence intervals, benchmarked against traditional scoring via ROC, precision-recall, and confusion matrices. The AI-enhanced method substantially outperformed traditional scoring (AUC = 0.94 vs. 0.72), with higher sensitivity (89% vs. 52%) and specificity (92% vs. 78%), reducing false negatives by 77%. Gains were consistent across evaluation metrics and robust to simulated age variation. These results demonstrate that multi-dimensional analysis of animal naming transforms a familiar task into a sensitive, interpretable digital biomarker for early MCI detection. Although based on simulated transcripts, the framework is designed for direct integration with automatic speech recognition and complements our AI-enhanced memory-list assessment. Prospective validation will establish clinical utility across care settings.