Quality enhancement in wearable devices of Photoplethysmographic signals affected by motion artifacts using Neural Networks

The removal of motion artifacts from PhotoPlethysmoGraphy (PPG) signals is a critical challenge in ensuring the accuracy and reliability of wearable devices designed for continuous monitoring of heart rate (HR) and other vital parameters. This model processes corrupted PPG signals in conjunction with 3-axis accelerometer data. The proposed approach effectively separates motion artifacts from the PPG signal without compromising the integrity of the physiological data, thereby improving the overall performance of wearable health monitoring devices. By integrating these inputs, the neural network generates a quality-enhanced PPG signal, achieving an average Pearson Correlation Coefficient (PCC) of 0.94 and an average Mean Absolute Error (MAE) of 0.17. The average error in heart rate (HR) calculation is 0.27 ± 0.15 bpm, with an average Peak Signal Score (PSS) of 0.01.