Publication

Abstract : Brain tumors pose huge challenges in diagnosis and treatment due to various reasons such as their aggressive nature, heterogeneity, and resistance to standard therapies. Accurate segmentation of brain tumors from multimodal magnetic resonance imaging (MRI) scans is crucial for treatment planning, disease progression tracking, and patient outcome prediction. This research proposes a novel U-Net-based deep learning method tailored for the BraTS 2020 challenge dataset and aims to improve brain tumor segmentation accuracy. The methodology integrates ensemble strategies, attention mechanisms, and residual connections to enhance the model's learning capacity and durability. Leveraging complementary data from various MRI modalities (T1, T2, T1c, FLAIR) using a multi-input, multi-output approach, the proposed method aims to better define tumor sub-regions. The integration of deep learning with radiomic characteristics from segmented tumors had been explored in this research for predicting overall survival and differentiating true recurrence from pseudoprogression. The study offers reliable testing across diverse imaging protocols and scanner configurations. The proposed method emphasizes the importance of multimodal MRI analysis in improving brain tumor diagnosis accuracy and contributes to the advancement of precise and trustworthy brain tumor segmentation algorithms and enhances overall patient care and treatment outcomes.