Medical Image Segmentation Using Enhanced Residual U-Net Architecture

Medical image segmentation is a crucial task in the field of medical imaging, and deep learning models have exhibited exceptional performance in recent years for segmentation purposes.In this paper, a refined network architecture of U-Net has been proposed, wherein residual units are included in U-Net to enhance the effectiveness of brain tumor segmentation.It constructs a deep learning model for the specific magnetic resonance imaging (MRI) segmentation task using the BraTS2020 dataset.The proposed enhanced model is designed by adding inner skip layers (residual connections) with fewer convolution layers to Allow the network to acquire knowledge of the residual mapping refers to the relationship between inputs layers and outputs layers instead of the direct mapping, consequently increasing the intersection over union (IoU).The results showed that after 100 epochs of training, the IoU of the proposed enhanced model is 0.910, while the model's accuracy is 0.968.In comparison, the original U-Net model achieved an Intersection over Union (IoU) score of 0.746 and an accuracy of 0.988 after 100 epochs.A comparison study was conducted with state-ofthe-art work to demonstrate the effectiveness of the proposed enhancement in improving the performance of deep learning models for MRI segmentation.The promising results clearly indicate the potential of this enhancement.