Reproduce “Deformed2Self: Self-Supervised Denoising for Dynamic Medical Imaging” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junshen Xu, Elfar Adalsteinsson
Deformed2Self introduces a self-supervised deep learning framework that integrates motion compensation with multi-frame information for denoising dynamic medical images. The framework effectively trains on a single noisy image series, demonstrating comparable or superior performance over existing state-of-the-art methods in improving image quality across diverse noise conditions.
Key insights
- Explicitly modeling and compensating for motion with STNs within a self-supervised framework effectively leverages temporal correlations in dynamic sequences for improved denoising.
- An initial coarse denoising step for individual frames enhances the accuracy of motion estimation, leading to better multi-frame information aggregation.
- The framework's ability to train on a single noisy image series, without requiring external clean ground-truth data, significantly improves its practical applicability in data-scarce medical scenarios.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.