Three-dimensional (3D) medical image segmentation involves partitioning a 3D medical image into distinct regions, each representing specific anatomical structures or classes. This paper introduces the plane attention module (PAM), targeting precise re...
Three-dimensional (3D) medical image segmentation involves partitioning a 3D medical image into distinct regions, each representing specific anatomical structures or classes. This paper introduces the plane attention module (PAM), targeting precise regions within 3D images by emphasizing plane-aware relationships and spatial information. Compatible with FCNN-based models like 3D U-Net, our architecture serves as an attention module. We evaluated its effectiveness utilizing the Kidney Tumor Segmentation Challenge 2023 dataset (KiTS 2023), specifically targeting tumor segmentation in abdominal computed tomography scans. Integrating the PAM into the 3D U-Net at encoder layer stage, we compared its performance with other attention modules and a robust baseline network. Results demonstrate that the 3D U-Net equipped with the PAM achieves an average 3.8-point Dice score improvement over the 3D U-Net baseline model.