A distributed model for automated diagnosis of whole-slide H&E stained prostate tissue images

Analysis of large amounts of medical images exceeds storage capacity and computation capability of a single workstation. Distributed computing employs a set of connected machines to solve a single problem by dividing it into number of solvable sub-problems. Analyzing and processing of large-scale medical images demand employing distributed architectures to overcome the limitations of memory space and execution time. Current analysis of digitized large-scale prostate tissue images depends on ordinary sequential techniques running on a single machine. This paper presents a proposed distributed model based on Hadoop framework for automated diagnosis of digitized large-scale H&E prostate tissue images to carry out segmentation, feature extraction, and classification tasks. The proposed model is based on partitioning input images into segments and distributing them across number of slaves to perform analysis task simultaneously. Analysis task aims at segmenting and labeling Regions of Interest (ROIs) in input images to extract initial features. Initial features are combined at master side to get final features for each input image. Finally, master node classifies images into the corresponding grade based on a grading system such as Gleason Grading system. The proposed distributed model would achieve high speed performance when applied in advanced medical applications.