Brain Tumor Segmentation of MRI Images Using Gabor Filter with Deep Learning Architectures

Abstract

Perfect segmentation of different region and subregions of gliomas from MRI scans has extremely high impact on treatment and diagnosis of brain tumors. .However, segmentation of the subregions is a difficult task because of the high irregular shape of these tumors. Numerous deep learning models have demonstrated their efficacy in a range of medical image segmentation tasks, which can be divided into two categories: intensity-based segmentation and shape-based segmentation. Most of them are based on the Segnet , U-Net and hybrid of these Models. In our study we use Gabor filter which is a linear filter used for edge detection and feature extraction. Gabor filter selects Phase, size, and frequency of the output. The features of the image are determined by Gabor filter by adaptively choosing size, frequency, and phase for each pixel. These features are compared to the features extracted by U-Net, ResNeXT50 and FPN. Common features are than used with FPN, U-Net and ResNExT50 with backbone U-Net for the segmentation of brain tumors. The Effectiveness of the proposed model has been calculated based on the IoU (intersection over union) score, Dice Score, and mean loss. The Dice Coefficient achieved are 95%, 90% and 91% for Gabor Filter x U-Net, Gabor x ResNext50 and U-Gabor x FPN respectively on the dataset of Brats’20. The outcome demonstrates that the suggested methodology is superior to other methodologies for segmenting MRI tumor brains.