A Benchmark Study of Deep Learning Algorithms for PPG Signal Processing

This study delves into extracting Photoplethysmography (PPG) signals from images using various neural network architectures. The research aims to assess the effectiveness of different models in capturing PPG values from snapshots of the dataset. Two training approaches were investigated: pretraining the model with only the fully connected part being trained or training the entire network. Results demonstrate that the models trained from scratch outperform their pretrained counterparts. Among the architectures, DenseNet121 shows the most promising results. The findings highlight the potential of utilising neural networks for PPG signal extraction, with applications ranging from surgical planning to personalised medical treatments. This research represents a significant stride in integrating advanced imaging techniques and neural networks in biomedical engineering.