Automated Organ Segmentation in 3D Biomedical Images Using Dual-Path 3D Attention U-ResNet

Deep learning algorithms have made considerable strides in recent years, and these advancements have the potential to significantly improve multi-organ segmentation in three-dimensional biomedical images. This would result in improved diagnostic accuracy and efficiency in medical imaging. DA-TransUNet is an innovative architecture that successfully combines information on several scales and enhances feature extraction by utilizing the DA-Block encoder and the Convolutional Block Attention Module (CBAM). The assessed propsoed model using a variety of datasets, one of which being the LiTS dataset. Finally, were able to achieve a Dice coefficient of 92.56% and a Hausdorff Distance (HD95) of 28.09 mm for liver images. DA-TransUNet surpasses existing state-of-the-art segmentation models across a variety of medical imaging datasets, as demonstrated by the results of the experiments, which reveal considerable increases in accuracy and robustness. The discoveries that have obtained contribute to the advancement of automated segmentation approaches in medical imaging, which in turn facilitates increased diagnostic capabilities.