Development of a neural network screening aid for diagnosing lower limb peripheral vascular disease from photoelectric plethysmography pulse waveforms

Abstract

An artificial neural network (ANN) was trained to classify photoelectric plethysmographic (PPG) pulse waveforms for the diagnosis of lower limb peripheral vascular disease (PVD). PPG pulses from the lower limbs, and pre- and post-exercise Doppler ultrasound ankle to brachial systolic blood pressure ratio measurements were obtained from patients referred to a vascular investigation laboratory. A single PPG pulse from the big toe of each leg was processed and normalized, and used as input data to the ANN. The ANN outputs represented the diagnostic classifications (normal, significant PVD and major PVD) and the ANN was trained with the ankle to brachial pressure indices (ABPI). The back-propagation learning algorithm was used to train the ANN for 500 epochs with a PPG training set of pulses from 100 legs. The results of this study indicate that a neural network can be trained to distinguish between PPG pulses from normal and diseased lower limb arteries.