Refined Attention Residual U-Net (RAResU-Net) for Efficient Lung Abnormality Segmentation

In the last few years, deep neural networks have been widely and effectively utilized in the segmentation of medical images, showcasing enhanced efficiency and accuracy. Within the realm of deep neural networks, U-Net models tailored for biomedical images have proven particularly successful. This study proposes a Refined Attention Res U-Net (RAResUNet) architecture for the automatic segmentation of abnormal lung nodules. The proposed method incorporates preprocessing techniques, and segmented using a Refined Attention Residual UNet (RARes-UNet). Notably, the encoder and decoder of RAResUNet were incorporated with a pre-trained ResNet and an attention block. In addition, the bottleneck was also augmented with ResNet and an attention block. Our testing, conducted on lung CT images from the database, revealed promising outcomes for the proposed approach is evaluated against state-of-the-art methods based on performance metrics. The proposed methods achieved results such as dice score of 89 %, a Jaccard score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 1. 5 \%}$</tex>, sensitivity of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 2 \%}$</tex>, and an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 1 \%}$</tex>.