Firefly Algorithm Assisted Segmentation of Tumor from Brain MRI using Tsallis Function and Markov Random Field

Abstract

Image segmentation plays a vital role in various medical applications for automated disease examination. In this paper, heuristic algorithm assisted approach is proposed to extract the tumor from a two dimensional (2D) magnetic resonance image. The proposed work is segregated into two processing regions, such as pre-processing and post-processing section.  In pre-processing, multi-level thresholding is applied for the 2D MR image using the Firefly Algorithm (FA) and Tsallis entropy function to cluster the similar image pixels based on an ideal intensity thresholds. For post-processing section, an image filter is initially considered to eliminate the skull region. The skull stripped image is then segmented into different partitions using Markov Random Field - Expectation Maximization (MRF-EM).  This procedure helps to attain three image segments, such as White Matter (WM), Gray Matter (GM) and tumor mass. The proposed method is tested on the MR images acquired using T1, T2 and Flair modalities. Standard image quality measures are considered to analyse the accuracy of pre-processing section. Further, the tumor mass is considered to examine the exactness in post-processing section. The efficacy of the proposed method is applied and validated validated using the BraTS 2D MRI dataset and achieved better values of Jaccard similarity coefficient and dice similarity co-efficient for all three modalities.