Brain Tumor Segmentation from MR Images using Customized U-net for a Smaller Dataset

In medical image analysis, deep learning has emerged as a powerful tool for solving complex tasks such as segmentation. This research presents an original approach using a customized U-net model for automatic brain tumor segmentation in Magnetic resonance imaging scans. The model's performance is thoroughly assessed using metrics like Mean Dice Similarity Coefficient, Sensitivity, Specificity, and Accuracy. The study evaluates the model's proficiency on 50 test images from in-house and established BRaTS 2020 datasets. The proposed U-net model showcases its superiority by achieving high Dice Similarity Coefficient scores, indicating a solid alignment with reference segmentations. It also demonstrates impressive sensitivity and specificity values, signifying its capacity to capture tumor regions and true negatives accurately. The results demonstrate that the proposed model effectively segments tumors on a relatively smaller dataset.