Developing Switching Mechanism to Address Challenging Subjects in Sleep Stage Scoring

Sleep stage scoring is a critical tool in sleep medicine but remains labor-intensive and challenging due to variability in subject attributes such as age. Existing automated methods show high overall accuracy but struggle with cases involving significant intersubject variability. This study proposes a novel approach that incorporates a Switching Mechanism based on selective desensitization into a CNN-based model. By dynamically adjusting the model’s weight vectors according to subject attributes, such as age, the proposed method tailors extracted features and scoring logic to individual characteristics. Experimental validation on the Cleveland Family Study dataset demonstrates that our method reduces the variability in scoring accuracy across subjects and achieves stable performance for cases with low baseline accuracy. These results highlight the potential of subject-specific adaptation in improving the robustness and clinical applicability of automated sleep stage scoring systems.