Summary

Researchers at the Paul Scherrer Institute PSI have developed an AI system called VISTACT that virtually stains tissue images from micro-computed tomography – like classical histological preparations. The system combines phase-contrast micro-CT with machine learning and was tested on lung tissue from patients with pulmonary hypertension. The study was published on June 17, 2026 in the Journal of the Royal Society Interface. The method enables non-destructive, three-dimensional tissue analysis without the preparation of thin sections.

People

Topics

  • Artificial Intelligence in Medicine
  • Pathological Diagnostics
  • Computed Tomography and Image Processing
  • Biomedical Research

Clarus Lead

The innovation addresses a century-old paradigm: Since Rudolf Virchow founded cellular pathology, disease diagnosis has been based on time-intensive analysis of stained tissue sections. Virtual 3D staining could fundamentally change this workflow – it promises speed, automatability, and spatial completeness that classical histology cannot offer. For tumor research, vascular diseases, and complex tissue architectures, this opens a new diagnostic window, even though the technology has not yet reached clinical routine maturity.

Detailed Summary

The VISTACT system is based on a two-stage technical innovation. First, phase-contrast micro-CT (PCµCT) utilizes not only differences in X-ray density but also additional information in the radiation itself – thereby achieving micrometer resolution for soft tissue in 3D, but delivering only grayscale images. Second, the team trained a specialized AI model (a "conditional Generative Adversarial Network") using pairs of real histological sections and corresponding CT images. The AI learned which microscopic patterns typically receive which staining – and could then automatically virtually stain new CT data.

A critical technical step was precise spatial registration: histological sections are only a few micrometers thick and can be distorted during sectioning. Lovric's team developed a multi-stage procedure that automatically recognizes the exact position of each section in the 3D dataset and matches it with histology data – significantly more precise than previous standard methods. In tests on lung tissue from patients with pulmonary hypertension (a disease characterized by pathological remodeling of lung vessels), the system was able to three-dimensionally map altered vascular regions. The AI plausibly differentiated between tissue components: blood in vessels appeared yellowish, collagen pink, lung surfaces gray to violet.

However, significant hurdles remain for clinical implementation. Phase-contrast imaging was performed at the TOMCAT beamline of the Swiss Light Source SLS – a large research facility at PSI. The data volumes are enormous, and the resolution often does not yet suffice to reliably display individual cell nuclei. Moreover, virtual histology remains a statistical reconstruction: the AI generates plausible predictions, not true histological information. Diagnostic routine quality is not yet achieved. However, the team emphasizes that the "proof of concept" has been established and the method is in principle transferable to various diseases.

Key Statements

  • Technological Breakthrough: First-time combination of phase-contrast micro-CT with AI-assisted virtual staining for 3D tissue analysis without sections or chemical staining
  • Clinical Potential: Accelerated research of disease biomarkers in tumors, vascular diseases, and complex tissue architectures
  • Maturity Limitation: Not yet routine-capable; currently requires synchrotron radiation and does not yet achieve diagnostic accuracy

Critical Questions

  1. Evidence/Data Quality: How large was the training dataset for the AI model, and how representative are the tissue samples used for the variability of real patient tissue?

  2. Validation Against Gold Standard: Were the virtual stainings systematically validated against conventional histology to exclude false-positive or false-negative diagnoses?

  3. Technical Scalability: Is the dependence on synchrotron radiation a fundamental obstacle, or could the method be adapted to conventional micro-CT devices?

  4. Statistical Limitations: Which specific tissue components or pathologies can the AI reliably depict, and where does it still fail?

  5. Clinical Workflow: How would this technology be integrated into the routine daily work of hospitals – as a screening tool before classical histology or as a replacement?

  6. Conflicts of Interest: Are there patent applications or commercialization plans that could affect availability for research?

  7. Side Effects/Risks: Could automated virtual staining lead to diagnostic errors if pathologists rely too heavily on AI output?

References

Primary Source: Virtual Tissue Staining in 3D – Paul Scherrer Institute PSI, 18.06.2026

Publication: Almagro-Pérez, C., Peruzzi, N., Galambos, C. et al. (2026). Histology-guided 3D virtual staining of microCT-imaged lung tissue via deep learning. Journal of the Royal Society Interface, 17.06.2026. DOI: 10.1098/rsif.2025.1186

Verification Status: ✓ 18.06.2026

This text was created with the support of an AI model. Editorial Responsibility: clarus.news | Fact-Check: 18.06.2026