Multimodal Fusion for OSA Detection Using Deep Neural Networks

This paper proposes a multimodal recognition method based on snoring sounds and respiratory tract structural variation videos for the auxiliary diagnosis of Obstructive Sleep Apnea Syndrome (OSA). It is important to note that snoring alone is not a definitive indicator of OSA; it is merely one potential symptom that should be considered in conjunction with additional information for a more accurate diagnosis. By integrating data from these two modalities, the diagnostic accuracy and robustness are significantly enhanced. Experimental results demonstrate that the proposed method outperforms single-modality approaches in key performance metrics, including accuracy and recall. This study provides a novel and effective solution for the diagnosis of OSA, offering new insights into the development of advanced diagnostic tools for sleep-related breathing disorders.