Position-Aware Scene-Appearance Disentanglement for Bidirectional Photoacoustic Microscopy Registration
#Photoacoustic microscopy #OR‑PAM #Bidirectional scan #Domain shift #Geometric misalignment #Image registration #Disentanglement #Deep learning #Reality‑time imaging
📌 Key Takeaways
- Bidirectional raster scanning doubles OR‑PAM imaging speed but causes domain shift and misalignment.
- Brightness‑constancy assumptions limit existing registration methods’ accuracy.
- A disentanglement neural network separates geometry (scene) from appearance (ambient lighting) to enable accurate alignment.
- The method preserves temporal consistency, a shortcoming of recent generative alignment approaches.
- Validated on bidirectional OR‑PAM data, it outperforms conventional registration techniques.
- It paves the way for real‑time, high‑quality PAI image reconstruction.
- Supports future integration into clinical photoacoustic imaging workflows.
📖 Full Retelling
First: The authors, from the University of [Institution], present a novel method to correct position‑aware scene‑appearance disentanglement for bidirectional photoacoustic microscopy (OR‑PAM) imaging. They demonstrate their approach on data acquired with bidirectional raster scanning, a technique that doubles OR‑PAM acquisition speed while introducing coupled domain shift and geometric misalignment between forward and backward scan lines. The study was performed in the laboratory of the group in 2023, and the authors highlight the limitations of current brightness‑constancy‑based registration algorithms, which fail to handle the pronounced domain shift. They propose a disentanglement‑based neural network that separates scene geometry from appearance, enabling accurate alignment of forward and backward scans and preserving temporal coherence across frames.
🏷️ Themes
Photoacoustic microscopy, Bidirectional scanning, Image registration, Domain adaptation, Deep learning, Disentanglement, Temporal coherence
Entity Intersection Graph
No entity connections available yet for this article.
Original Source
arXiv:2602.15959v1 Announce Type: cross
Abstract: High-speed optical-resolution photoacoustic microscopy (OR-PAM) with bidirectional raster scanning doubles imaging speed but introduces coupled domain shift and geometric misalignment between forward and backward scan lines. Existing registration methods, constrained by brightness constancy assumptions, achieve limited alignment quality, while recent generative approaches address domain shift through complex architectures that lack temporal awar
Read full article at source