Multi-modal MRI Image Segmentation of Brain Tumors for Multi-class Segmentation
By: Justin Luong, Seyoung Kim
Department: Computer Science
Faculty Advisor: Dr. Ilmi Yoon
The project uses a CNN called UNET to perform multi-class segmentation on the BraTS dataset, which consists of MRI images of brain tumors. The network is trained on the training set and evaluated on the validation set using the Dice loss function. Decoders are used to preserve high-resolution features, and the output is a 2D segmentation mask. The performance is evaluated using various metrics and compared to state-of-the-art methods. The resulting segmentation maps can aid clinicians in the diagnosis and treatment of brain tumors.