New Survey Proposes Taxonomy for Controllable 3D CT Generation
A new survey paper on arXiv (2608.09992) introduces a classification system aimed at organizing the rapidly expanding field of conditional 3D Computed Tomography (CT) generation. It addresses the challenge of assessing different methods, which is crucial for clinical applications such as data augmentation, safe data sharing, and simulating specific anatomical or pathological scenarios. The framework categorizes existing research into three key aspects: the type of external knowledge (K), the knowledge integration approach (I), and the generative architecture (A). This taxonomy helps to clarify design choices and facilitates meaningful comparisons. Targeted at medical imaging and generative deep learning experts, the survey provides a thorough guide for creating controllable generation methods.
Key facts
- The paper is a survey on conditional 3D CT generation.
- It proposes a conditioning-centric taxonomy with three dimensions: knowledge type (K), integration paradigm (I), and generative architecture (A).
- The taxonomy defines a design space (K x I x A).
- The survey addresses the difficulty of comparing diverse approaches in the field.
- Clinical applications include data augmentation, privacy-preserving data sharing, and simulation of anatomical or pathological scenarios.
- The paper is available on arXiv with ID 2608.09992.
- The announcement type is 'cross'.
- The survey aims to clarify fundamental design choices in 3D CT generation.
Entities
Institutions
- arXiv