Development of an enhanced U-Net model for brain tumor segmentation with optimized architecture

Abstract

The paper aims at an enhanced deep learning-based brain tumor segmentation model of MRI images. The input MRI images are pre-processed by the filtering and contrast enhancement techniques, followed by patch extraction. In the pre-processed image, the segmentation of the images is processed by the Enhanced U-Net model; in turn the architecture is optimally tuned by the improved Coyote Optimization Algorithm (COA) called Adaptive Searched Coyote Optimization Algorithm (AS-COA). The hyper-parameters of the U-Net architecture like batch size and the epoch count are optimized by solving the objective function as dice coefficient maximization. The post-processing step includes the morphological closing model for extracting the relevant region of interest with maximum accuracy. The proposed model finally estimates the dice scores for improving whole tumor, tumor and core tumor. Hence, the suggested method's experimental findings demonstrate that it is capable of accurately segmenting brain tumors, as evidenced by comparisons with conventionalalgorithms.