Researchers propose new taxonomy for AI-generated 3D CT scans

A team of researchers from Politecnico di Milano has published a comprehensive survey that introduces a conditioning-centric taxonomy for 3D Computed Tomography (CT) generation. The framework, proposed by Francesca Pia Panaccione, Eugenio Lomurno, and Matteo Matteucci, organizes existing approaches along three orthogonal dimensions: external knowledge type (K), knowledge integration paradigm (I), and generative architecture (A). This creates an explicit design space (K × I × A) that provides a unified perspective on prior work in controllable 3D CT generation.

The survey addresses a critical gap in the field where the rapid expansion of research has made systematic comparison difficult. By categorizing methods through this taxonomy, the researchers identified dominant trends and design patterns while highlighting underexplored areas. The framework is particularly valuable for clinical applications including data augmentation, privacy-preserving data sharing, and simulating specific anatomical or pathological scenarios. The work has been accepted to IJCAI-ECAI 2026's Survey Track.