DAMamba-UNet3D: A Parameter-Efficient Mamba State Space U-Net with Dynamic Adaptive Scan for 3D Medical Image Segmentation

2026-07-21

AI Summary

This work addresses the challenge in 3D medical image segmentation of simultaneously achieving effective global context modeling and parameter efficiency, compounded by the absence of data-driven dynamic scanning mechanisms. To this end, we propose DAMamba, a hybrid U-Net architecture that integrates Mamba state-space modules with triplanar dynamic adaptive scanning (3D-DAS) exclusively into the encoder while retaining convolutional operations elsewhere, enabling efficient fusion of global and local features. The method introduces an encoder-specific DAS embedding strategy that substantially reduces model parameters while enhancing segmentation accuracy. Evaluated on BraTS 2020 using five-fold cross-validation, our base model (5.3M parameters) achieves a Dice score of 0.815, and the larger variant, DAMamba-L (70M parameters), reaches 0.829—outperforming SegMamba with a 13-fold improvement in parameter efficiency.

Abstract

We propose parameter-efficient SSM-based U-Net architectures for 3D medical image segmentation. Convolutional U-Nets afford O(n) local mixing per layer but lack explicit global context; transformers provide global reasoning at O(n^2) cost in sequence length $n$. State-space models (SSMs), such as Mamba, offer $O(n)$ global propagation per block. Yet, existing medical SSM segmenters rely on fixed scan patterns and large parameter budgets. Dynamic Adaptive Scan (DAS), which learns data-dependent reordering before selective scan, has not been applied to medical imaging or extended to 3D volumes. We propose DAMamba-UNet3D, a hybrid encoder-decoder that integrates tri-plane 3D-DAS blocks at encoder stages E2-E4 while retaining convolutions elsewhere (~5.3M parameters). On BraTS 2020 five-fold cross-validation, DAMamba-UNet3D achieves mean Dice 0.815+/-0.013 (full-volume per-case evaluation) at ~13x lower parameter cost than SegMamba (0.824+\-0.014, ~70M). At comparable scale, DAMamba-L (~70M), a wide DAS-native variant with encoder-only DAMamba and a convolutional bottleneck, reaches 0.829+\-0.012, surpassing retrained SegMamba by 0.5pt. Component ablations show that encoder-only DAS placement is critical as bottleneck and decoder SSM blocks lower Dice. Together, the results suggest that learned tri-plane DAS in a hybrid U-Net is competitive with, and under our large-scale design may improve upon, SegMamba's fixed Tri-orientated Mamba (ToM) scanning on BraTS 2020. Code: https://github.com/marafathussain/DAMamba-UNet3D.

Problem

• 3D medical image segmentation
• state-space models
• dynamic adaptive scan
• parameter efficiency
• U-Net

Innovation

• Dynamic Adaptive Scan
• State Space Model
• 3D Medical Image Segmentation
• Parameter-Efficient U-Net
• Tri-plane Scanning