Deep learning-based detection of cessation of breathing in pre-term infants

2026-06-22 · Source: Takara TLDR - Daily AI Papers · Field: Health & Wellbeing — Medical Devices & Health Technology, Clinical Care & Medical Practice, Health & Medical Research · Depth: Expert, quick

Summary

Deep learning models can reliably detect apnoea-related Cessation Of BrEathing (COBE) events in pre-term infants using routinely monitored Neonatal Intensive Care Unit (NICU) physiological signals. Researchers evaluated shallow convolutional neural networks, residual networks, and ConvNeXt architectures on approximately 430 hours of impedance pneumography (IP), electrocardiography (ECG), and photoplethysmography (PPG) recordings from 24 pre-term infants. The dataset included 346 COBE and 608 non-COBE events. Detection performance was primarily influenced by signal modality, not architectural complexity. Unimodal IP-based models achieved 86.8-88.0% balanced accuracy, outperforming ECG-derived (62.6-69.7%) and PPG-derived (65.1-66.4%) surrogates. The best model, a ConvNeXt combining IP and PPG, reached 88.7% balanced accuracy and an F1 score of 0.75, demonstrating deep learning's potential for enhanced neonatal monitoring.

Key takeaway

For AI Scientists developing neonatal monitoring systems, prioritize impedance pneumography (IP) as the primary signal for apnoea detection. Your models, even simpler ones, will achieve high balanced accuracy (86.8-88.0%) with IP alone. While multimodal fusion with PPG can offer a slight improvement to 88.7% balanced accuracy, focus first on robust IP signal processing, as signal modality significantly outweighs architectural complexity in performance.

Key insights

Deep learning reliably detects infant apnoea using routine NICU signals, with signal modality being key.

Principles

• Signal modality impacts detection more than architecture.
• Impedance pneumography is superior for apnoea detection.
• Multimodal fusion offers modest performance gains.

Method

Evaluated CNN, ResNets, and ConvNeXt on IP, ECG, and PPG signals from 24 pre-term infants, using a dataset of 346 COBE and 608 non-COBE events.

In practice

• Integrate IP as primary apnoea detection signal.
• Consider ConvNeXt for multimodal physiological data.
• Prioritize signal quality over model complexity.

Topics

• Apnoea of Prematurity
• Deep Learning
• Neonatal Intensive Care Unit
• Impedance Pneumography
• ConvNeXt
• Physiological Monitoring

Best for: AI Scientist, Research Scientist

Related on AIssential

• Hybrid deep learning model for multimodal vocal and lung signal analysis in health monitoring — A hybrid deep learning model effectively integrates vocal and lung signal analysis for accurate disease detection.
• Contrast-Informed Augmentation and Domain-Adversarial Training for Adult-to-Neonatal MR Reconstruction Generalization — Contrast-informed augmentation and domain-adversarial training enhance deep learning MR reconstruction generalization from adult to neonatal data.
• AI stethoscope trial highlights the perils of implementation gaps — AI diagnostic tools require careful integration into clinical workflows to achieve their full potential.
• Evaluation of EEG Foundation Models for Event-Based Burst-Suppression Detection in ICU — EEG Foundation Models show promise for robust, patient-agnostic burst-suppression detection in ICU settings, even with limited data.
• StreamPPG: Low-Latency rPPG Estimation via Consistent Privileged Learning — StreamPPG leverages consistent privileged learning to enable low-latency, accurate remote photoplethysmography estimation suitable for real-time edge device deployment.
• Pretrained Neural Audio Models for Asthma Detection from Voice and Speech — Pretrained neural audio models can effectively detect asthma from mobile-recorded speech, outperforming scratch training.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Takara TLDR - Daily AI Papers.