An Efficient U-Net based Model for Low Grade Glioma Segmentation in MRI Images

Low-grade gliomas (LGG) are brain tumors that grow slowly and originate from glial cells. Detecting these tumors at an early stage is crucial, and Magnetic Resonance Imaging (MRI) serves as an effective screening method for this purpose. Precise identification of tumor borders in MRI images is crucial for effective treatment planning. This study employs an efficient U-net based model for segmenting brain tumors in Magnetic Resonance images. Through meticulous data pre-processing and augmentation techniques, the proposed model is trained to accurately delineate intricate tumor boundaries. The model's performance was evaluated using the Intersection Over Union (IoU) and Dice coefficient metrics on a publicly available Low- Grade Glioma (LGG) segmentation dataset. Experimentation with various backbone architectures reveals that employing ResNet-50 outperforms VGG16 and VGG19. Furthermore, com- parison with state-of-the-art methods shows that the proposed model surpasses existing works in LGG segmentation.