Cristina Almagro-Pérez held her breath as the first full 3D map of a diseased lung emerged in vivid, histology-grade color—no scalpel, no stains, no slides. At the Paul Scherrer Institute in Switzerland, she and physicist Goran Lovric had just proven that decades-old methods of tissue analysis could leap into a new dimension. For over a century, pathologists have relied on slicing tissue into paper-thin sections, staining them with dyes, and examining them under microscopes—a painstaking, two-dimensional ritual rooted in Rudolf Virchow’s 19th-century cell theory. Now, thanks to a breakthrough AI platform called VISTACT, that process may soon be transformed.
VISTACT—Virtual Staining of micro-Computed Tomography—combines high-resolution phase-contrast micro-CT (PCµCT) with deep learning to produce colorized, 3D tissue images that mimic traditional histology. Unlike standard CT, which detects only X-ray density, PCµCT captures subtle phase shifts in X-rays, revealing soft tissue structures at micrometer resolution. But until now, these images remained in grayscale, lacking the critical color cues pathologists depend on: blue-violet nuclei, pink collagen, dark elastic fibers. VISTACT bridges that gap by training an AI model on paired datasets of real histological slides and their corresponding CT scans. The result? A machine that learns to “translate” grayscale 3D data into richly detailed, color-annotated tissue maps.
The team’s innovation hinges on precision. Histological sections, often just a few micrometers thick, can warp during preparation, making alignment with 3D scans error-prone. Lovric’s group developed a multistage registration process that automatically matches each tissue slice to its exact location in the CT volume—achieving unprecedented spatial accuracy. Then, using a conditional generative adversarial network (cGAN), the system applies virtual stains not as flat color fills, but as biologically plausible renderings: collagen glows pink, blood in capillaries appears yellowish, and lung surfaces shift from gray to violet.
When tested on lung tissue from patients with pulmonary hypertension—a condition marked by abnormal blood vessel remodeling—the technique successfully mapped pathological changes in 3D, offering insights impossible to capture in 2D sections. "We have shown for the first time that a CT-based virtual stain can deliver results similar to conventional laboratory histology," Lovric says. "This could open up a wealth of clinical and scientific applications."
The implications are profound. Virtual staining could accelerate diagnoses, reduce lab costs, and preserve entire tissue architectures for repeated analysis. As AI and imaging converge, VISTACT points toward a future where disease is no longer studied slice by slice—but as a living, three-dimensional story.