New AI Approach to Improve Surgical Imaging
Posted on 03 Feb 2026
Surgeons often rely on visual judgment to distinguish between healthy and diseased tissue or to avoid damaging critical structures during procedures. This assessment is subjective and limited by what the human eye can perceive, increasing the risk of error in complex surgeries. Advanced imaging techniques such as hyperspectral imaging can reveal hidden information like blood flow and oxygenation, but translating this data into real-time clinical guidance remains challenging. Researchers have now developed an artificial intelligence (AI) approach that enables accurate interpretation of such data without relying on large amounts of human training images.
The approach, developed by researchers at the German Cancer Research Center (DKFZ, Heidelberg, Baden-Württemberg, Germany), in collaboration with Heidelberg University Hospital (UKHD, Heidelberg, Baden-Württemberg, Germany) and Mannheim University Medical Center (UMM, Mannheim, Baden-Württemberg, Germany), is called “xeno-learning”. The method allows AI systems to learn from hyperspectral medical images collected in animal studies and transfer that knowledge to human tissue. This strategy bypasses the ethical and practical barriers associated with collecting large, annotated image datasets from patients.
Hyperspectral cameras capture far more information than conventional imaging by recording detailed spectral signatures related to tissue composition and physiology. To make this data clinically useful, AI algorithms must be trained to recognize meaningful patterns. Instead of learning absolute color values, the xeno-learning approach trains neural networks to recognize relative changes associated with pathological processes such as impaired circulation, allowing the model to generalize across species.
In the study, researchers analyzed more than 13,000 hyperspectral images from humans, pigs, and rats. Conventional AI models trained on animal data performed poorly when applied to human tissue. In contrast, the xeno-learning model successfully transferred knowledge by focusing on disease-related patterns rather than species-specific signatures. The findings demonstrate that animal imaging data can be used to train AI systems capable of accurate human tissue assessment.
The researchers believe this approach could significantly improve surgical safety and precision, particularly in situations where human training data is scarce or unavailable. By enabling AI-assisted spectral imaging without patient-derived datasets, xeno-learning could accelerate the adoption of advanced imaging technologies in the operating room. To support rapid translation, the team has released the program code and pre-trained models for use by other researchers.
“Xeno-learning enables the use of spectral imaging even where human data is lacking,” said Jan Sellner, one of the two lead authors of the study.
“This is an important step toward making surgical procedures safer and more precise in the future,” added Alexander Studier-Fischer, who led the clinical aspects of the project.