Automated Identification of Cyclic Alternating Patterns of Sleep Using Fusion of VGG16 and Vision Transformer

Sleep plays a crucial role in human health and significantly impacting physical and mental well-being. An automated system is crucially needed to identify Cyclic Alternating Pattern (CAP) phases A and B, which are pivotal in assessing sleep depth and stability, enhancing the understanding of sleep and diagnosing sleep disorders. Our study addresses this critical need by proposing an automated classification system capable of accurately distinguishing CAP phases A and B. Leveraging advanced machine learning techniques, including the fusion of VGG-16 and Vision Transformer models, along with preprocessing methods, we aim to develop a robust system for CAP phase identification. By utilizing the PhysioNet dataset, encompassing a diverse range of subjects, from healthy individuals to those with various sleep disorders, such as Insomnia, Narcolepsy, Nocturnal Frontal Lobe Epilepsy (NFLE) and Periodic Limb Movement (PLM), our approach seeks to provide comprehensive insights into sleep patterns and disorders. Furthermore, our results demonstrate promising accuracies, with notable performance improvements over individual architectures. Specifically, accuracies of 94.29% for healthy subjects, 96.02% for Narcolepsy, 90.49% for Insomnia, 90.38% for PLM and 89.73% for NFLE were achieved. These findings highlight the effectiveness of the proposed approach in accurately identifying CAP phases and diagnosing sleep disorders, thus contributing to advancements in automated sleep analysis and healthcare.