Automatic analysis of the histology microscopy

[jamesavery]

In order to evaluate correctness of bone-implant contact (BIC) calculations from the SRmuCT images, we want to compare to the microscopy images of the corresponding histologocal samples.

These are found in threadripper00:/mnt/data/MAXIBONE/Goats/tomograms/histology/ in multiresolution .ndpi format (full resolution is 50k x 50k, i.e. about 10x the SRmuCT-resolution, but effective resoution is lower due to optical effects). An example of how to load the files is found in inspect_images.ipynb in the same directory.

Challenges: The tissue is dyed with blue, but unevenly so. Segmentation needs a bit of thought/preprocessing - pixel-value based segmentation would need help. Striped marks from the diamond saw also makes gradient-based segmentation difficult.

Matching with the histology is also useful for other analyses, and in particular: We can match the histological samples to their positions in the 5mu-tomograms. Can features visible in the optical images (but invisible in the X-ray tomograms) be recovered and extended to 3D volumes?