Multimodal AI-approach for the automatic screening of cardiovascular diseases based on nocturnal physiological signals

• Published in: NPJ cardiovascular health Vol. 2; no. 1; p. 15
• Main Authors: Kim, Youngtae, Jang, Tae Gwan, Park, So Yeon, Park, Ha Young, Lee, Ji Ae, Oyun-Erdene, Tumenbat, Kim, Sang-Ha, Park, Young Jun, Cho, Sung Pil, Park, Junghwan, Kang, Dongwon, Urtnasan, Erdenebayar
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 06.05.2025; Nature Publishing Group
• Subjects: 692/4019; 692/699; Artificial intelligence; Biomedical and Life Sciences; Cardiac Imaging; Cardiac Surgery; Cardiology; Cardiovascular Biology; Cardiovascular disease; Cell Biology; Chronic illnesses; Electrocardiography; Life Sciences; Medicine/Public Health
• ISSN: 2948-2836; 2948-2836
• DOI: 10.1038/s44325-025-00051-z

This study proposes a multimodal AI algorithm called the SleepCVD-Net to automatically screen CVDs based on nocturnal physiological recordings. We designed and implemented a multimodal AI algorithm, SleepCVD-Net, which utilizes three-mode deep neural networks to process input signals—single-lead electrocardiography (ECG), Airflow, and oxygen saturation (SpO 2 ). Nocturnal physiological recordings were extracted from 194 subjects (80 controls and 114 subjects with CVD) in the Sleep Heart Health Study database. The proposed SleepCVD-Net model demonstrated good performance, achieving a mean accuracy of 97.55% on the test set. The F1-scores were 97.97%, 96.35%, 97.79%, and 97.49% for the control, stroke, angina, and congestive heart failure groups, respectively. The results indicate the potential for the automatic screening of CVDs based on nocturnal physiological signals. Furthermore, the SleepCVD-Net can serve as a valuable tool for monitoring cardiac activity during sleep in inpatient, outpatient, and home healthcare settings.