Brain Tumor Detection and Classification using Deep Learning Techniques

Background: Brain tumor localization and segmentation from magnetic resonance imaging (MRI) are challenging yet crucial tasks in medical analysis. Recent approaches have utilized multiple MRI modalities, such as T1, T1c, T2, and FLAIR, to capture unique tumor characteristics. Despite promising results on datasets like BRATS 2018, many existing methods suffer from complexity and overfitting issues, requiring extensive training and testing time. To address these challenges, we propose a two-step approach for flexible and efficient brain tumor segmentation. Firstly, we introduce a preprocessing method that focuses on a small region of interest within each image slice, reducing computation time and mitigating overfitting. Secondly, we present a Cascade Convolutional Neural Network (C-CNN) designed to extract both local and global features through separate routes. Additionally, we introduce a novel Distance-Wise Attention (DWA) mechanism to enhance segmentation accuracy by considering the spatial relationship between the tumor center and surrounding brain tissue. Our proposed method achieves competitive results on the BRATS 2018 dataset, with mean whole tumor, enhancing tumor, and tumor core dice scores of 0.9203, 0.9113, and 0.8726, respectively. Quantitative and qualitative assessments further validate the effectiveness of the proposed model. The proposed approach offers a flexible and effective solution for brain tumor segmentation, demonstrating improvements over state-of-the-art models. By incorporating preprocessing techniques, a specialized network architecture, and the DWA mechanism, our method addresses key challenges in MRI-based tumor analysis, paving the way for more accurate and efficient medical image processing.