MOSAIC: Modality-Specific Adaptation for Incremental Continual Learning in Parkinson's Disease Gait Assessment

MOSAIC is a novel continual learning framework designed for gait assessment in Parkinson's disease, addressing challenges in modality-incremental learning with heterogeneous sensor data. Key innovations include the Modality-Specific Warm-Up to counteract the Toxic Teacher phenomenon, a statistics-decoupled MSBN architecture for isolating sensor statistics, and a curriculum-guided repulsive objective for enhancing plasticity while retaining legacy knowledge. Experimental results on three multimodal datasets demonstrate improved performance and reduced forgetting, making MOSAIC significant for practitioners working with incremental learning in clinical settings.