Heart Rate Estimation from Neck Photoplethysmography using FFT-Based Scoring and a Shallow Neural Network

Heart rate (HR) is one of the most important vital signs to monitor. Constantly monitoring HR makes it important to have an easy-to-use system that is able to achieve clinically acceptable measurement accuracy. Depending on the monitoring device's ultimate intended use, the sensing modality as well as body location can be suboptimal for cardiac signal acquisition. In this work, neck photoplethysmography (PPG) signals were used to estimate heart rate using a novel method of FFT-Based scoring coupled with a shallow neural network. This method was able to achieve an average RMSE value of 3.13 ± 4.66, MAE of 1.96 ± 3.38, and error STD of 2.41 ± 3.23 when testing on all data without exclusions, and a competitive average RMSE value of 1.55±1.43, MAE of 0.83±0.86, and error STD of 1.31±1.16 when excluding skewing outlier participants' data.