BRAIN TUMOR SEGMENTATION AND TUMOR VOLUME ESTIMATION FROM MRI USING U-NET, ATTENTION UNET, DEEP V3+

Abstract

Brain tumors are among the most critical neurological disorders, where early and accurate detection plays a vital role in effective diagnosis and treatment planning. Magnetic Resonance Imaging (MRI) is widely used for brain tumor detection due to its ability to provide detailed visualization of brain tissues. However, manual segmentation of tumor regions from MRI scans is time-consuming, labor-intensive, and prone to human error. To overcome these challenges, automated deep learningbased segmentation techniques have emerged as efficient and reliable solutions. This study focuses on brain tumor segmentation and tumor volume estimation using advanced convolutional neural network architectures, namely U-Net, Attention U-Net, and DeepLabV3+. These models are trained on MRI datasets to automatically detect and segment tumor regions. U-Net utilizes an encoder–decoder structure for effective feature extraction, while Attention U-Net enhances segmentation accuracy by emphasizing relevant tumor regions through attention mechanisms. DeepLabV3+ further improves performance by employing atrous convolution and spatial pyramid pooling to capture multi-scale contextual information. Following segmentation, the extracted tumor region is used to estimate tumor volume, which is a critical factor in clinical assessment and treatment planning. The models are evaluated using performance metrics such as Dice coefficient, Intersection over Union (IoU), and accuracy.