The low five-year survival rate for lung cancer underscores the importance of early detection. A key component of this process is the accurate segmentation of pulmonary nodules from CT scans to quantify their characteristics. While deep learning models have advanced this field, Transformer-based architectures face limitations due to their high computational complexity with high-resolution medical images. This paper introduces Lung-Mamba, a deep learning model for lung nodule segmentation that combines a U-Net framework with the recently proposed Mamba architecture. Mamba utilizes Selective State Spaces to model long-range dependencies with linear complexity, offering an efficient alternative to Transformers. The proposed architecture integrates Mamba layers into a U-Net to capture both local features and global context. Evaluated on the LIDC-IDRI dataset, using 12,465 nodules for training and 3117 for testing, Lung-Mamba achieves a Dice score of 96.48%. This result positions the model as an effective and computationally efficient method for medical image segmentation, demonstrating the benefit of integrating state-space models into established convolutional frameworks. • Lung-Mamba : A U-Net and Mamba-based model for pulmonary nodule segmentation. • Integration of Selective State Space (SSM) blocks to capture long-range dependencies. • Achieves a Dice score of 96.48% on the LIDC-IDRI dataset. • Outperforms baseline models in segmentation accuracy. • Provides a computationally efficient architecture for a high-resolution imaging task.
Gayap et al. (Fri,) studied this question.