Brain Tumour Segmentation in MRI Scans using Enhanced 3D U-Net Model

Magnetic Resonance Imaging (MRI) is a critical tool for diagnosing brain tumours, offering high-resolution insights into soft tissues. However, manual interpretation of MRI scans is time-consuming, prone to errors, and limited by the complexity of tumour shape and imaging condition. Automated segmentation addresses these limitations by delivering consistent and accurate tumour boundaries, enabling effective treatment planning. Traditional methods, such as thresholding and region growing, often struggle with irregular tumour shapes and indistinct boundaries. Machine learning and deep learning models, particularly 3D U-Net architectures, have shown promise in overcoming these challenges. This study introduces an enhanced 3D U-Net model for brain tumour segmentation in MRI scans, incorporating architectural improvements, multi-scale contextual learning, and optimised hyperparameters. The proposed model was evaluated using the BraTS 2020 dataset, the model achieved a mean Intersection Over Union (IoU) of 0.8185 and a Dice Similarity Coeffi-cient (DSC) of 0.6852, demonstrating its effectiveness in managing complex tumour shapes and diverse imaging conditions. These results highlight the model's potential to improve segmentation reliability and clinical decision-making. The proposed approach represents a significant step toward scalable and precise brain tumour segmentation solutions for real-world applications.