Biometrics have a great deal of potential in healthcare applications, most notably authentication for medical record privacy and fraud prevention. In this paper, we examine, for the first time, non-fiducial feature extraction for photo-plethysmography (PPG) based authentication. PPG signals have unique identity properties for human authentication, and are becoming easier to capture by emerging IoT sensors such as MaxFast. Different machine learning techniques are used to compare non-fiducial and fiducial feature extractions. Our experimental results show that 99.84% accuracy with EER of 1.31% can be achieved based on non-fiducial feature extraction.