Prompt-guided dual-path UNet with Mamba for medical image segmentation

Convolutional Neural Networks (CNNs) and Transformers are widely adopted in UNet-based architectures for medical image segmentation. However, CNNs struggle to model long-range dependencies, while Transformers suffer from quadratic computational complexity. Recently, Mamba, a State Space Model, has shown promise due to its ability to capture long-range interactions with linear computational complexity. Despite the emergence of several Mamba-based methods, their architectures still exhibit limited perceptual awareness of the input images and tend to emphasize global context while neglecting essential local details. To address these challenges, we propose a novel prompt-guided dual-path CNN-Mamba UNet, termed PGM-UNet, for medical image segmentation. Specifically, we introduce a prompt-guided residual mamba module (PGRM) that leverages a contrastive language-image pre-training (CLIP) model to adaptively extract dynamic visual prompts from the input images. This mechanism effectively guides Mamba in capturing global information with enhanced perceptual awareness. Additionally, we design a local-global information fusion network, which comprises a local information extraction module, the PGRM, and a multi-focus attention fusion module, to effectively integrate both local details and global context. Furthermore, inspired by Kolmogorov-Arnold Networks (KANs), we develop a multi-scale information extraction module to capture richer contextual information without altering feature resolution. Extensive experiments on the ISIC-2018, CVC-ClinicDB, DIAS, DRIVE, and Spleen datasets demonstrate that PGM-UNet outperforms state-of-the-art approaches across multiple medical image segmentation tasks. Our code is available at https://github.com/shaoleizhang01/PGM-UNet .