Three-dimensional X-ray histology offers a non-invasive alternative to conventional 2D histology, enabling volumetric imaging of biological tissues without physical sectioning or chemical staining. However, the intrinsic greyscale contrast of X-ray tomography limits its biochemical specificity compared with traditional histological stains. In this study, we extend deep-learning-based virtual staining to the X-ray domain via cross-modality image translation to generate artificially stained slices directly from synchrotron radiation microtomography (µCT) scans. Using over 50 co-registered pairs of µCT and toluidine blue-stained histology from bone-implant samples, we trained a modified CycleGAN network tailored for limited paired data. Whole-slide histology images were downsampled to the CT voxel size, with on-the-fly data augmentation for patch-based training. The model incorporates pixelwise supervision and greyscale consistency losses, enabling histologically realistic colour outputs while preserving structural detail. Results outperformed Pix2Pix and standard CycleGAN baselines across metrics of structural similarity, perceptual fidelity, and peak signal-to-noise ratio. Once trained, the model can be applied to full µCT volumes to produce virtually stained 3D datasets that enhance interpretability without additional sample preparation. This work introduces virtual staining to 3D X-ray imaging, which may provide a scalable route for chemically informative, label-free tissue characterization in biomedical research.
Irvine et al. (Fri,) studied this question.