Multi-frame Restoration for High-rate Lissajous Confocal Laser Endomicroscopy
A lightweight AI fills missing pixels in ultra-fast endomicroscopy videos for clear tissue imaging.
Lissajous confocal laser endomicroscopy (CLE) enables high-speed in vivo optical biopsy, but at high frame rates its resonant scanning pattern leaves most pixels unvisited, producing structured holes. In a new arXiv preprint, researchers from multiple Korean institutions present the first benchmark for high-rate Lissajous CLE, consisting of low-quality video clips paired with high-quality reference images. Those references are wide field-of-view mosaics created from stabilized, slow-scan frames of the same tissue, enabling temporally aligned supervision.
To solve the restoration problem, the team proposes MIRA, a lightweight recurrent framework that iteratively aggregates temporal context through feature reuse and displacement alignment. Experiments show MIRA outperforms both lightweight and high-complexity baselines in restoration quality while maintaining favorable computational efficiency suitable for clinical deployment. The work marks a key step toward practical, real-time high-resolution microscopy for handheld diagnostic scenarios.
- First benchmark for high-rate Lissajous CLE with paired low-quality video and high-quality reference mosaics.
- MIRA uses recurrent feature reuse and displacement alignment to fill unvisited pixels across frames.
- Outperforms lightweight and high-complexity baselines with computational efficiency suited for clinical use.
Why It Matters
Enables real-time, high-resolution optical biopsy without motion artifacts, advancing in vivo diagnostics.