Automated Detection and Segmentation of Heterogeneous Brain Tumor Regions Via Transformer Based Deeplabv3+

A tumor is one of the most dangerous diseases that affect any body part, where tumors of the brain are the most serious because of their important place. They pose an important risk to life and are an important cause of death because of their complexity and treatment challenges. This paper uses the transformer deeplabv3+ model, starting with employ data pre-processing techniques, including augmentation and resizing given images. The Gated Shape Convolutional Neural Network (GSCNN) model is trained using a training set as a base model, expanded with the pre-trained ResNet50 weight, and the predictions are then fed into the DeepLabv3+ model. The model's performance is assessed based on its F1 score, accuracy, sensitivity, precision, and specificity. As per the study's outcomes, our suggested model enhances performance and can achieve 98.87% accuracy, 98.84% sensitivity (recall), 99.49% specificity, 98.83% precision, and a 98.83% F1-score. Compared to earlier studies, our model produced beneficial results, utilizing the dataset BraTS2020.