Brain Tumor Segmentation Using Deep Learning (BraTS2020)

Manual brain tumor segmentation is time-consuming and subjective. Our approach demonstrates how deep learning can assist radiologists by providing accurate, consistent, and fast tumor delineation, supporting better diagnosis and treatment planning.

Technical Highlights:

Dataset: BraTS 2020

Architecture: UNet-based deep learning model

Preprocessing: Intensity normalization, CLAHE, background-only slice removal

Evaluation Metrics: Dice Score, IoU, Accuracy, PSNR, SSIM

Key Results:

Accuracy: 99.61%

Dice Score: 0.917

IoU: 0.852

Average PSNR: 37.05 dB

Average SSIM (tumor region): 0.994

These results show a high level of agreement between predicted segmentation masks and expert annotations, with strong robustness across different tumor sizes and MRI modalities.