Brain Tumor Segmentation with FPN-Based EfficientNet and XAI

Brain tumors are complex and dangerous conditions that require accurate diagnosis for effective treatment. While the Magnetic Resonance Imaging (MRI) is a crucial diagnostic tool, the process of interpreting and evaluating MRI is time-consuming and requires knowledge from the experts. Developing and using machine learning methods to predict brain tumors can speed up diagnosis, reduce wait times, and could improve accuracy. This study proposes using deep learning methods, e.g., the EfficientNet model combined with the Feature Pyramid Network (FPN) to segment brain tumors in reality. For validation the proposed approach, we trained model on the BraTS 2020 dataset, achieving good performance on the test and evaluation sets. The proposed method demonstrated an average IoU accuracy of 0.9083 and 0.8878 and an average Dice accuracy of 0.9336 and 0.9303 on the test and evaluation sets, respectively. Moreover, we have used Grad-CAM for visualization the results to explain and understand more about the prediction. Results show that the proposed approach could be used in practice for helping the doctors in medical domain.