Breaking01 / AI
MICRO benchmark pairs retinal images across surgery phases
Researchers proposed a one-step learning method that aligns retinal images using similarities in their blood vessel structures.

Researchers introduced MICRO, a benchmark comprising 1,365 retinal image pairs for alignment before, during and after eye surgery, in a paper abstract published on arXiv on 11 Nov.
The abstract describes images captured using different imaging methods, with standardised annotations. Registration is the process of aligning blood vessel structures between images.
The researchers also proposed OFS2Reg, a one-step registration learning method that uses similarities between vessel structures. It applies a frequency-based feature representation to highlight those structures.
According to the abstract, the method learns similarities between vessel structures to refine how reference points correspond across images. It is designed to align vessels at pixel level across different imaging methods.
The researchers contrasted this approach with existing two-step methods that first identify vessel regions and then detect reference points. They described those methods as heavily dependent on the quality of the vessel segmentation—the separation of vessel regions from the rest of an image.
The abstract does not disclose registration accuracy or numerical performance comparisons with existing methods.