Balancing Efficiency and Restoration: Lightweight Mamba-Based Model for CT Metal Artifact Reduction

MARMamba addresses a high-value niche in medical imaging by applying Mamba (SSM) architectures to the Metal Artifact Reduction (MAR) problem. Its primary value proposition is achieving performance comparable to heavy Transformers but with linear scaling and lower memory overhead, specifically designed to be sinogram-independent (operating directly on images). With 10 forks but 0 stars in just 9 days, the project shows immediate interest from the academic/research community (likely reviewers or researchers in the MAR field), which is a stronger signal than general 'star' counts for specialized domains. Defensibility is currently low (4) because it's a research-centric reference implementation without a proprietary dataset or clinical integration moat. While frontier labs (OpenAI/Google) are unlikely to build specific CT artifact tools, the real threat comes from medical imaging giants like GE Healthcare, Siemens, or Philips, who could easily integrate similar Mamba-based modules into their proprietary reconstruction pipelines. The novelty lies in the specific adaptation of Mamba blocks to handle the non-local nature of CT artifacts while preserving anatomical structural integrity.