Enhanced Deep Learning Model Performance in 3D Multimodal Brain Tumor Segmentation with Gabor Filter

Medical image segmentation plays a crucial role in identifying and analyzing anatomical structures in medical images. This requires an accurate medical segmentation tool to delineate and quantitatively analyze the target regions, diagnose any abnormality, and assist in treatment planning. Deep learning approaches have emerged as a promising solution for automating medical segmentation. However, challenges arise when dealing with the complex shapes and spatial variations of some target regions, especially in 3D MRI scans. To deal with such transformations, specific techniques are required to properly analyze and preprocess the dataset and perform image filtering to provide better features for improved prediction performance of deep learning architectures. This study focuses on improving brain tumor segmentation in multimodal 3D MRI images. We observed significant improvements in multimodal brain tumor segmentation results (accuracy, IoU, and mIoU) using an optimized 3D Gabor filter, which helps extract meaningful features. Multichannel input images were preprocessed to remove noise and select an appropriate resolution to reduce computational complexity. An improvement in mean Intersection over Union (mIoU) from 0.714 to 0.804 and accuracy from 0.982 to 0.991 were achieved, which shows a major improvement. This work contributes to the field of medical image segmentation by offering an improved and efficient approach for brain tumor analysis in 3D MRI scans, potentially aiding in diagnosis and treatment planning.