MLCDT: fine-grained multi-task learning for enhanced cognitive assessment in the clock drawing test

Artificial Intelligence (AI) is increasingly used in cognitive health assessments, with the Clock Drawing Test (CDT) being an effective cognitive evaluation tool. However, the complexity of CDT image structures, high subjectivity, and the lack of specialized cognitive health assessment datasets for specific populations pose significant challenges for feature learning and model construction using this method. To address these issues, we propose a fine-grained multi-task learning approach (MLCDT) for AI-assisted diagnosis of cognitive health using CDT. MLCDT integrates image pre-training models with a multi-task learning framework to capture fine-grained features of CDT images and constructs a final diagnostic support model through scientifically designed tasks. Experiments using real data from cognitive health assessments in a neurology department at a hospital validate the effectiveness of MLCDT in handling fine-grained tasks and aiding cognitive disorder assessments.