GPU fuzzy c-means algorithm implementations: performance analysis on medical image segmentation

Abstract

Image segmentation in the medical imagery such as MRI, is an essential step to the sensitive analysis of human tissues lesions with the objective to improve the partition of different parts of the image according to their specific characteristics. Fuzzy c-means (FCM) is one of the widely used algorithms in literature regarding image segmentation. Indeed, it offers performances to the precision level in many medical fields of applications. However, the main limitation of FCM algorithm is time consuming during the image segmentation by clustering. In order to improve and to reduce the time delay of image data processing, we implemented three methods inspired from the FCM on GPU GT 740 m by using the CUDA environment. This latter is well adapted to the new architectures of processing, and its sequential migration towards the parallel approach through the SIMD architecture as GPU cards solves this time constraint. Furthermore, we have improved, via the two current developed implementations methods, the speed up of the processing acquisition in comparison with the reference methods. The efficiency evaluation such as strengths and weaknesses of each implemented method will be evaluated on medical images segmentation according to the size of the modelled brain tumours.