PCM-based Architecture for Compressed Sensing on Skin Ulcers Images and Automatic Classification with CNN Neural Network

Phase Change Memory (PCM) represents a tech-nology that exploits the reversible phase transition of a chalco-genide material to create nanoscale memory components, which can be used for the development of brain-inspired computing approaches. These PCM devices have been examined both as non-volatile storage-class memory and as computing elements for in-memory and neuromorphic computing applications. It is also known that PCM exhibits several characteristics of a memristive device. In this study, we consider a PCM array as an encoder within a system for applying Compressed Sensing (CS) to images of skin ulcers. We then use a decoding strategy that compensates for the non-linearity of PCM devices through an iterative optimization approach. The quality of image reconstruction was evaluated by classifying the images using a Convolutional Neural Network (CNN) according to Wound Bed Preparation (WBP) severity scale, which is used in clinical practice for the assessment of skin lesions. The effectiveness of the image compression and reconstruction was demonstrated by comparing the automatic classification performance on before and after CS images.