Uni-COAL: A unified framework for cross-modality synthesis and super-resolution of MR images

作     者：Song, Zhiyun Qi, Zengxin Wang, Xin Zhao, Xiangyu Shen, Zhenrong Wang, Sheng Fei, Manman Wang, Zhe Zang, Di Chen, Dongdong Yao, Linlin Liu, Mengjun Wang, Qian Wu, Xuehai Zhang, Lichi 

作者机构：Shanghai Jiao Tong Univ Natl Engn Res Ctr Adv Magnet Resonance Technol Dia Sch Biomed Engn Shanghai 200030 Peoples R China Fudan Univ Huashan Hosp Shanghai Med Coll Dept Neurosurg Shanghai 200040 Peoples R China Natl Ctr Neurol Disorders Shanghai 200040 Peoples R China Shanghai Key Lab Brain Funct & Restorat & Neural R Shanghai 200040 Peoples R China Fudan Univ State Key Lab Med Neurobiol Shanghai 200040 Peoples R China Fudan Univ Frontiers Ctr Brain Sci Sch Basic Med Sci MOE Shanghai 200040 Peoples R China Fudan Univ Sch Basic Med Sci Shanghai 200040 Peoples R China Fudan Univ Inst Brain Sci Shanghai 200040 Peoples R China ShanghaiTech Univ Sch Biomed Engn Shanghai 201210 Peoples R China 

出 版 物：《EXPERT SYSTEMS WITH APPLICATIONS》 (Expert Sys Appl)

年 卷 期：2025年第270卷

核心收录：

学科分类：1201[管理学-管理科学与工程(可授管理学、工学学位)] 0808[工学-电气工程] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

基　　金：National Natural Science Founda-tion of China 

主　　题：Cross-modality synthesis Super-resolution Magnetic resonance imaging 

摘      要：Magnetic resonance imaging (MRI) enhancement techniques, such as cross-modality synthesis (CMS), super- resolution (SR), and their combination (CMSR), are crucial to improve the quality of images produced by simplified scanning protocols. However, existing approaches are typically tailored to specific tasks, limiting their flexibility and generalizability to handle complex clinical scenarios. Moreover, these MRI enhancement methods often treat alias frequencies improperly, resulting in suboptimal detail restoration performance. In this paper, we propose a Unified CO-modulated ALias-free framework ( Uni-COAL ) to accomplish the aforementioned tasks with a single network, so that the interaction of these tasks can be established and the resources for training and deploying the models can be greatly reduced. Specifically, the co-modulation design of the image- conditioned and stochastic attribute representations ensures the task consistency between CMS and SR, while simultaneously accommodating arbitrary combinations of modalities and thickness. The generator of Uni-COAL is also designed to be alias-free based on the Shannon-Nyquist signal processing framework, ensuring effective suppression of alias frequencies. Additionally, we leverage the semantic prior of the Segment Anything Model (SAM) to guide Uni-COAL, ensuring a more authentic preservation of anatomical structures during synthesis. Experiments on four datasets demonstrate that Uni-COAL outperforms the alternatives in various combinations of CMS, SR, and CMSR tasks, which highlights its superiority and flexibility to wide-range applications. Our codes are publicly available at https://***/zhiyuns/UniCOAL.