Year: 2026
Type: article
Abstract: For pediatric patients with respiratory insufficiency, long-term non-invasive ventilation (LT-NIV) represents an essential treatment. Despite significant technological progress, the commercial built-in ventilator software has structural limitations: its automatic scoring algorithms are typically calibrated for adult respiratory patterns, failing to detect the high respiratory rate variability inherent to children. As a result, clinicians must perform a manual data review to identify episodes of obstructive sleep apnea. While Machine Learning (ML) has demonstrated significant potential in identifying these events, most studies focus on adults and employ multi-channel polysomnography data. Addressing this, we investigated automated methods for sleep apnea detection using only a single-channel respiratory signal from built-in ventilator software. Therefore, we propose a novel hybrid approach combining a Bidirectional Long Short-Term Memory (BiLSTM) network with a Feature Engineering-based ML ensemble. Experimental results, validated through a Leave-One-Out strategy, demonstrate that the hybrid model can identify sleep apnea events despite the high variability in pediatric respiratory rates and the significant class imbalance within the dataset. The findings provide encouraging evidence for the potential use of the proposed approach as a decision-support tool for automated analysis of respiratory signals in pediatric LT-NIV. However, prospective validation on larger and more diverse pediatric cohorts will be necessary before clinical deployment.
Language: English
Subfield: Physiology (medical)
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $1,870
OpenAlex ID: W7223200910