MedVAR: Towards Scalable and Efficient Medical Image Generation via Next-scale Autoregressive Prediction

Published in arXiv preprint arXiv:2602.14512, 2026

MedVAR introduces a next-scale autoregressive framework for efficient and scalable medical image synthesis. It supports coarse-to-fine generation and is trained with a harmonized large-scale CT/MRI dataset across multiple anatomical regions.

Recommended citation: Zhicheng He, Yunpeng Zhao, Junde Wu, Ziwei Niu, Zijun Li, Lanfen Lin, Yueming Jin. (2026). "MedVAR: Towards Scalable and Efficient Medical Image Generation via Next-scale Autoregressive Prediction." arXiv preprint arXiv:2602.14512.
Download Paper