<p>Accurate segmentation of pulmonary nodules has become increasingly critical for the early detection of lung diseases. However, conventional lung nodule segmentation models often suffer from poor real-time performance and high computational resource demands. To address these challenges, we propose a novel Adaptive Feature Shuffle Lightweight V-Net (AFS-LVNet). First, we replace standard convolution operations with a Multi-dimensional Convolution Shuffle Module (MCSM), which leverages dilated grouped convolutions and channel splitting to expand the receptive field while reducing computational complexity. This design improves the model’s ability to capture global context and fine-grained details. Second, we introduce an Adaptive Receptive Field Module (ARFM) to enhance multi-scale feature extraction of pulmonary nodules with minimal additional cost. In place of traditional up-downsampling, we employ a Pixel Shuffle Module (PSM) to reduce artifacts and preserve spatial information. Additionally, a Feature Shuffle Attention (FSA) module is integrated to dynamically recalibrate feature responses by combining channel and spatial attention mechanisms, thereby improving segmentation precision. Experimental results on the LUNA16 public dataset demonstrate that AFS-LVNet outperforms the original V-Net, achieving a 3.75% improvement in Dice Similarity Coefficient (DSC), a 36.88% reduction in FLOPs, and only 89.28% of the parameter count of V-Net. Compared to existing lightweight approaches, AFS-LVNet shows significant improvements in both segmentation accuracy and computational efficiency, highlighting its potential for deployment in resource-constrained medical applications.</p>