CUDA parallel programming technology application for analysis of big biomedical data based on computation of effectiveness features

Abstract

This paper proposes the technology for large biomedical data analysis based on CUDA computation. The technology was used to analyze a large set of fundus images used for diabetic retinopathy automatic diagnostics. A high-performance algorithm that calculates effective textural characteristics for medical image analysis has been developed. During the automatic image diagnostics, the following classes were distinguished: thin vessels, thick vessels, exudates and a healthy area. The study of the mentioned algorithm efficiency was conducted with 500x500-1000x1000 pixels images using a square 12x12 dimension window. The acceleration relationship between the developed algorithm and various data sizes was demonstrated. The study showed that the algorithm effectiveness can be affected by certain characteristics of the image, e.g. its clarity, shape of exudate zone, variability of blood vessels, and optic disc location.