MobileMamba-UNet boosts low-dose CT reconstruction with wavelet-enhanced efficiency

MobileMamba-UNet, a lightweight hybrid model, outperformed prior deep learning methods for low-dose CT reconstruction on Mayo datasets, delivering better image quality with lower memory use and faster inference (retrospective testing).

• Design: Integrates MobileMamba backbone, wavelet-enhanced Mamba mechanism, and multi-scale U-Net for efficient long-range modeling.
• Results: Consistently outperforms CNN- and Transformer-based methods on Mayo-2016 and 2020 LDCT datasets; exact quantitative improvements not specified in abstract.
• Limitation: Evaluated only on public retrospective datasets; no prospective clinical or external validation reported.

RadPigeon summaries are original and for information only. They are not clinical advice.