Lung Nodule Segmentation Integrating Convolutional Attention in V-Net: A Comparative Benchmark with U-Net and V-Net

Accurate segmentation of pulmonary nodules on computed tomography (CT) scans is crucial for lung cancer early detection and treatment planning. Deep learning algorithms, par-ticularly convolutional neural networks (CNNs) like U-Net and V-Net, have exhibited significant performance in medical image segmentation tasks. Nevertheless, these architectures may find it challenging to consistently capture subtle contextual information and complex boundary details of small lung nodules. In this study, we present an enhanced V-Net model that incorporates the Convolutional Block Attention Module (CBAM) to facilitate improved feature representation through adaptively enhancing both spatial and channel-wise information. A comprehensive benchmark analysis is conducted between the baseline U-Net, the original V-Net, and the introduced CBAM-enhanced V-Net (CBAM-VNet) on a well-prepared lung CT dataset with expert-annotated nodule masks. The experimental results indicate that CBAM-VNet consistently surpasses the baseline models, particularly in accurately delineating nodules characterized by complex textures and indistinct boundaries. The proposed CBAM-VNet achieves the highest segmentation performance, yielding a Dice score of 0.65 and an accuracy of 99.81 %. These findings highlight the effectiveness of incorporating attention mechanisms into 2D V-Net models to improve the accuracy and robustness of automated lung nodule segmentation.