SRGAN Reconstructs Ultrafast Synthetic MRI Quantitative Maps

Researchers at Beijing Friendship Hospital and Beihang University report in JMIR Med Inform (2026) that a superresolution generative adversarial network (SRGAN) can reconstruct ultrafast synthetic MRI into whole-brain T1, T2, and proton-density maps closely matching routine scans. In a prospective study of 151 healthy adults and seven patients, SRGAN outputs showed strong correlation (T1 R²=0.98; T2 R²=0.97; PD R²=0.99) and halved acquisition time while maintaining diagnostic image quality.