ResNet Encoder Based UNet for Brain Tumour Segmentation

In the past few, the Convolutional Neural Network has been ruling the domain of image processing. Medical Imaging plays a critical role in the clinical diagnosis and treatment planning. This has significantly reached several milestones in medical image segmentation since the development of U-Net architecture. In this paper, we propose a novel Residual Encoder with UNet architecture enhanced with ASPP for skip connections for Brain tumour segmentation, thereby improving the feature extraction and mitigating the vanishing gradients problem during the training. By using the strengths of residual learning, the encoders capture the deep hierarchical features of the image, while the decoder of the UNet facilitates the reconstruction of spatial features of the tumour regions. This is crucial as it improved the segmentation of the brain tumour more effectively and aids in diagnosis, planning and monitoring the treatment effectiveness. Unlike the traditional UNet, our approach residual learning and ASPP module to strengthen the model's ability to capture diverse tumour shapes and sizes, providing robust segmentation results. In this proposed model we incorporate the IoU (Intersection over Union) to demonstrate the performance of the model's efficiency and achieving the better segmentation results.