Researchers develop AI to align X-ray and optical microscopy images
New method achieves 93% accuracy in aligning X-ray fluorescence with optical images
Researchers from Argonne National Laboratory and Northwestern University have developed an AI-assisted method to precisely align X-ray fluorescence (XRF) microscopy with optical microscopy images. The team, led by Xiangyu Yin, introduced 'acquisition geometry-assisted whole-group localization,' which uses the spatial relationships between XRF tiles recorded during scanning to improve alignment accuracy.
In controlled experiments, their method achieved a GroupIoU (intersection-over-union) score of 0.931, compared to 0.000 for independent tile placement. The approach also demonstrated robustness across different similarity metrics, with mutual information (MI) performing nearly as well as normalized cross-correlation (NCC). In multiscale case studies, using coarse XRF survey scans to guide fine-scale alignment improved GroupIoU from 0.694 to 0.856, highlighting its scalability for complex imaging workflows.
- Method by Xiangyu Yin et al. achieves 93.1% localization accuracy vs. 0% for independent placement in controlled tests
- Uses acquisition geometry (spatial relationships between XRF tiles) to constrain alignment, improving GroupIoU from 0.694 to 0.856 in multiscale cases
- Works independently of similarity metrics (NCC or MI) and scales with coarse-to-fine XRF survey scans
Why It Matters
Enables seamless integration of elemental distribution and morphology data in medical imaging, improving diagnostic accuracy and research precision.