Newborn heart rate estimation using respiratory signals: Applying curriculum learning concept via transfer learning approach

• Introduced a novel deep learning framework for neonates. • Single device estimates heart rates from respiratory signals. • Hybrid CNN-LSTM model minimizes sensor use in NICUs. • Curriculum learning improves heart rate estimation accuracy. • Method distinguishes between low and high-quality signals. Monitoring neonatal vital signs is essential in neonatal intensive care units (NICUs). Current systems often require multiple sensors attached to neonates, risking skin injuries and hindering parent-infant bonding. This study introduces a deep learning framework that estimates neonatal heart rates from respiratory inductance plethysmography, aiming to reduce the need for multiple sensors by analyzing respiratory signals through a single device. The framework combines a convolutional neural network (CNN) with a long short-term memory (LSTM) model, implementing curriculum learning by training the model on high-quality signals before fine-tuning with low-quality signals. Patient-level split cross-validation was performed to avoid data leakage, ensuring model robustness and generalizability. Results demonstrate significant improvements with curriculum learning across all models, with the hybrid CNN-LSTM model achieving a 7.33 % reduction in mean absolute error (MAE) from 10.64 bpm to 9.85 bpm, outperforming standalone 1D-CNN and LSTM models. For low-quality signals, the hybrid model yielded the lowest MAE (9.13 bpm), while LSTM performed best on high-quality signals (7.96 bpm). These findings highlight the technical innovation and practical implications of this approach, offering a cost-effective, skin-friendly alternative for NICU monitoring. Beyond reducing sensor dependency, this framework supports vital neonatal-maternal bonding and minimizes risks associated with skin-attached sensors, demonstrating the feasibility and promise of respiratory signal analysis for critical monitoring applications in neonatal healthcare technology.