High Performance Implementation of Fuzzy C-Means and Watershed Algorithms for MRI Segmentation

Journal of Advances in Computer Research

مقاله 2، دوره 10، شماره 1 - شماره پیاپی 35، اردیبهشت 2019، صفحه 11-24   

نوع مقاله: Original Manuscript

نویسندگان

Farnaz Hoseini^* ؛ Ghader Mortezaie Dekahi

Department of Computer Engineering, Faculty of Engineering, Shahriar Institute of Higher Education, Astara, Iran

چکیده

Image segmentation is one of the most common steps in digital image processing. The area many image segmentation algorithms (e.g., thresholding, edge detection, and region growing) employed for classifying a digital image into different segments. In this connection, finding a suitable algorithm for medical image segmentation is a challenging task due to mainly the noise, low contrast, and steep light variations of medical images. Due to the inherently parallel nature of image segmentation algorithms, they suit well for implementation on a Graphics Processing Unit (GPU). The main goal of this paper is to improve the performance of fuzzy c-means clustering through the parallel implementation of this algorithm. Although fuzzy c-means clustering is an important iterative clustering algorithm, it is computationally intensive and uses the same data between the iterations. The center of the clusters changes in each iteration, which requires a considerable amount of time for large data sets. The parallel fuzzy c-means clustering is implemented by applying pipeline parallelism on GPU. The experimental results show that the performance is improved up to 23.35x. Next, the watershed algorithm is applied to the final segmentation. In this paper using parallel fuzzy c-means clustering and computations we have attained competing results with other papers. The implementation results on the BRATS2015 show that the accuracy of diagnosis in Dice Similarity Coefficient metric 97/33% is obtained. This improvement is achieved using enhancing edges and reducing noises in images.

کلیدواژه‌ها

Parallel Fuzzy C-Means Clustering؛ Watershed Algorithm؛ FCM؛ CUDA؛ GPU