What question did this study set out to answer?

The aim is to develop a reliable AI model, BCGNet, for effective sleep monitoring using non-invasive methods.

July 5, 2026npj Digital Medicine0 citationsOpen Access

BCGNet: an AI model trained on 600 K hours of sleep data for a novel under-pillow contactless monitoring device

SCShigeng ChenGriffith University XCXuesong ChenChinese Academy of Medical Sciences & Peking Union Medical College WHWeijun HuangShanghai Sixth People's Hospital

Key Points

The aim is to develop a reliable AI model, BCGNet, for effective sleep monitoring using non-invasive methods.
Developed BCGNet using transfer learning pre-trained on 580,865 hours of PSG data and fine-tuned on 15,081 hours of BCG recordings.
Validated performance across multiple validation cohorts for sleep staging, AHI estimation, and sleep architecture quantification.
Assess the model's generalizability on diverse external datasets.
Achieved an F1 score of 0.710−0.817 for 4-class sleep staging.
Estimated Apnea-Hypopnea Index with a Pearson’s r > 0.95.
Maintained strong performance on short daytime naps and across various external datasets.

Abstract

Abstract Sleep profoundly impacts health, yet current gold-standard Polysomnography (PSG) is constrained by cost, discomfort, and limited scalability for longitudinal monitoring. Ballistocardiography (BCG) offers a non-invasive and user-friendly alternative but often lacks the precision needed for reliable real-world applications. To address this gap, we propose BCGNet, a two-stage transfer learning model that is first pre-trained on 580,865 h of PSG and then fine-tuned and validated on 15,081 h of BCG (total 595,946 h of recordings). Across multiple validation cohorts, BCGNet achieves strong performance in 4-class sleep staging (F1: 0.710−0.817), Apnea-Hypopnea Index (AHI3%) estimation (Pearson’s r > 0.95), and robust quantification of sleep continuity and architecture (ICC and Pearson’s r generally >0.8). Notably, BCGNet maintains strong performance even on short daytime naps and demonstrates excellent generalizability across diverse external datasets. Deployed as a portable, contactless sleep tracking mat, BCGNet represents a major step towards scalable, user-friendly solutions for longitudinal home sleep monitoring, with important implications for population screening and personalized sleep medicine.

Ask AI

Helpful

Bookmark

View Full Paper