BRAIN TUMOR SEGMENTATION USING DUAL FUSION ENHANCED DEFORMABLE CONVOLUTIONAL NETWORK AND ARTIFICIAL LEMSTAR OPTIMIZATION

Abstract

Brain tumors are highly aggressive and pose a serious threat to life expectancy across varied demographic groups. Detecting them early allows for timely treatment, which greatly improves survival chances. However, segmentation is complicated by similar imaging intensities, irregular tumor shapes, and unclear boundaries. Hence, this paper presents a new method using the Dual Fusion with Enhanced Deformable Convolution Network (DFEDC) and Artificial LemStar Optimization Algorithm (ALStarOA) for segmenting brain tumors using Magnetic Resonance Images (MRI). Firstly, the MRI image is given to image denoising using Spatial Domain Filtering (SDF). Then, DFEDC is utilized to perform brain tumor region segmentation, and the loss function of DFEDC is modified by using the Exponential Baikal Loss. DFEDC is trained by exploiting ALStarOA, which is developed by combining StarFish Optimization Algorithm (SFOA) and Artificial Lemming Algorithm (ALA). Experimental analysis shows that the proposed methodology outperforms with a higher segmentation accuracy of 97.467%, dice coefficient of 97.777%, and Intersection Over Union (IOU) of 98.279%.