Brain tumor detection is a critical task in medical imaging, requiring accurate and timely identification for effective treatment planning. This study focuses on the segmentation of FLAIR (Fluid-Attenuated Inversion Recovery) abnormalities in brain MRI images using a deep learning-based approach. Leveraging U-Net, a well-established convolutional neural network architecture for biomedical image segmentation, we implemented a model in PyTorch to achieve precise tumor segmentation. The U-Net model was trained and evaluated on publicly available datasets, with preprocessing steps including normalization and augmentation to enhance the model’s generalizability. Our methodology emphasizes automated segmentation to alleviate the challenges associated with manual annotation, which is often time-consuming and subject to variability. The U-Net architecture, known for its encoder-decoder structure, effectively extracts critical spatial features while retaining high-resolution output for accurate segmentation. A Dice coefficient-based loss function was employed to optimize performance and assess segmentation accuracy. Experimental results demonstrate the model’s effectiveness in accurately segmenting FLAIR abnormalities, achieving a high Dice similarity coefficient and Intersection over Union (IoU) score across test datasets. These findings underscore the potential of deep learning approaches in advancing brain tumor diagnosis by providing clinicians with reliable tools for automated analysis. In conclusion, this study highlights the feasibility of using U-Net in PyTorch for brain tumor detection and segmentation, presenting a robust solution to a challenging medical imaging problem. Future research will explore the integration of multi-modal imaging data and advanced post-processing techniques to further enhance segmentation accuracy and clinical applicability.