Virtual Staining Technique Creates Histology Images from CT Data

Yet routine pathology remains anchored in labor‑intensive, two‑dimensional staining of delicate sections, which limits spatial context and consumes time. These constraints can impede comprehensive visualization of disease architecture needed to guide diagnosis and research. To help address this challenge, researchers have now developed a three‑dimensional virtual staining approach that reproduces familiar histology directly from computed tomography data.

Developed at the Paul Scherrer Institute (PSI), the platform—Virtual Staining of Micro‑Computed Tomography (VISTACT)—combines high‑resolution phase‑contrast micro‑computed tomography (PCµCT) with machine learning to generate colorized images that mirror conventional histologic stains. The system translates gray‑scale volumetric scans into tissue‑specific color contrasts recognizable to pathologists. It aims to preserve intact samples while revealing fine anatomic structures throughout a specimen in three dimensions.

The team trained a specialized AI model using paired datasets that linked true histologic sections with their corresponding micro‑CT slices. A multi‑stage registration workflow precisely located each thin section within the three‑dimensional dataset and compensated for distortions introduced during sectioning and mounting. Using a conditional generative adversarial network for image‑to‑image translation, the model produced virtual stains from gray‑scale CT input.

Proof of concept was demonstrated in lung tissue from individuals with pulmonary hypertension. The approach plausibly differentiated tissue components, rendering blood in fine vessels yellowish, collagen pink, and lung surfaces gray to violet, while preserving reference cues such as blue‑violet nuclei and dark elastic fibers. Investigators reported that CT‑based virtual stains delivered results similar to laboratory histology and enabled three‑dimensional mapping of remodeled vascular regions. The work was published in the Journal of The Royal Society Interface on June 17, 2026.

Important limitations remain before routine clinical deployment. Phase‑contrast imaging was performed at the TOMCAT beamline of the Swiss Light Source, generating extremely large datasets. Resolution frequently did not permit reliable depiction of individual cell nuclei, and the virtual staining represents statistical predictions based on training data rather than direct chemical staining.

Even with these constraints, the method establishes proof of concept for non‑destructive three‑dimensional pathology. The researchers note its potential applicability to tumors, vascular lesions, and complex tissue architectures. They add that the approach could accelerate biomarker research and, over time, broaden diagnostic options as imaging access and resolution improve.

Related Links
Paul Scherrer Institute