PPG-Based Respiration Monitoring Using CycleGAN

This paper presents a deep learning-based approach for estimating respiratory rate (RR) from Photoplethysmogram (PPG) signals, a non-invasive method that measures blood volume variations in peripheral vessels. Conventional RR measurement techniques, such as impedance pneumography and capnography, require specialized hardware, manual calibration, and patient-specific adjustments, limiting scalability and ease of deployment. A new propose a Cycle Generative Adversarial Network (CycleGAN) framework to reconstruct respiratory waveforms directly from raw PPG data, enabling automated and accurate RR estimation without the need for traditional respiratory sensors. The model employs paired generator and discriminator networks to map between PPG and respiratory domains while enforcing cycle consistency for high fidelity signal reconstruction. Pre-processing steps include normalization, down sampling, and segmentation of PPG signals into fixed-length windows for model training. Using the BIDMC PPG and Respiration Dataset, our approach achieved a Mean Absolute Error (MAE) of 0.52 and a Respiration Rate Error of 4.24, demonstrating robust performance and adaptability across datasets. This methodology offers strong potential for integration into wearable health monitoring devices, telemedicine platforms, and other real-time non-invasive diagnostic systems, with future work exploring real-time deployment and extension to other physiological signal translations.