DSA-Net : A Novel Approach for Brain Tumor Segmentation

A brain tumor develops due to an abnormal proliferation of cells within the brain tissue. Early detection is crucial for patient safety, and Magnetic Resonance Imaging (MRI) scans are commonly used for diagnosing brain tumors. However, due to the diverse shapes and locations of tumors in the brain, accurate segmentation of tumors in MRI images remains a challenge for physicians. The variability in tumor appearance, often influenced by factors such as edema, necrosis, and surrounding healthy tissues, makes precise delineation difficult. Precise segmentation is essential for identifying the tumor and providing appropriate treatment. It allows clinicians to measure the tumor’s size and progression over time, enabling personalized treatment adjustments based on the individual patient’s response. This study introduces a novel deep learning technique called Depthwise Separable Attention Network (DSA-Net) for automatic tumor segmentation. The proposed DSA-Net model demonstrates a significantly higher Dice coefficient compared to existing models, indicating its effectiveness in brain tumor segmentation.