Non-Invasive Glucose Level Monitoring from PPG using a Hybrid CNN-GRU Deep Learning Network

Summary

This arXiv paper presents a hybrid CNN-GRU deep learning network for non-invasive blood glucose estimation from raw PPG signals. The motivation is that PPG contains information about blood optical properties that correlate with glucose concentration, but extracting this signal is difficult due to noise, motion artifacts, and inter-individual variability. The CNN component extracts local morphological features from PPG waveforms, while the GRU captures longer-range temporal dependencies across multiple heartbeat cycles.