MPSoC FPGA Implementation of Algorithms of Machine Learning for Clinical Applications Using High-Level Design Methodology

Abstract

This paper presents the design of an FPGA-accelerated application for skin cancer detection which uses both hyperspectral imaging and a k-means algorithm. The accelerator is designed employing 3 FPGA kernels. The first 2 kernels filter and normalize the hyperspectral image. Then, the last kernel runs k-means to segment the image into three different regions according to the distribution of the lesion. This application is developed following the HLS methodology, implemented as an embedded system in MPSoC, and runs under Linux OS. FPGA acceleration will improve the application's throughput and energy efficiency significantly when compared to pure software execution.