Abstract
In contemporary healthcare, precise segmentation of medical pictures is crucial for diagnosis and therapy planning. By automating the hitherto time-consuming human segmentation process, deep learning techniques like CNNs, U-Nets, and Transformers have completely changed this sector. Using deep learning algorithms, this work offers a novel method for brain tumor MRI picture segmentation that tackles the difficulties encountered in contemporary healthcare. We propose an enhanced U-Net model explicitly tailored for this purpose, aiming to improve segmentation precision. Our methodology incorporates feature augmentation techniques during the preprocessing phase, including Contrast Limited Adaptive Histogram Equalization (CLAHE), Multi-histogram Equalization (MHE), and Modified Biorthogonal Histogram Equalization (MBOBHE). Furthermore, we modify the U-Net architecture with a customized layered design to optimize segmentation outcomes. A Convolutional Neural Network (CNN) is employed for post-processing, applying additional convolutional layers to refine the segmentation results. The model demonstrates impressive performance metrics, achieving a recall of 93.66%, accuracy of 97.79%, F-score of 93.15%, and precision of 92.66%. The Dice coefficient's training and validation curves indicate strong generalization capability, with training at approximately 93% and validation at 84%. Despite the model's strengths, challenges remain, particularly concerning false positives, highlighting the need for ongoing refinement. By helping radiologists with precise diagnosis and treatment planning, this work provides the foundation for more
accurate and efficient medical image segmentation approaches, which will have a substantial influence on healthcare.