A Complex‐Valued Super‐Resolution Method for Functional MRI

ABSTRACT The goal is to develop and evaluate a complex‐valued super‐resolution MRI framework for recovering high‐resolution functional and structural brain images, with the goal of mitigating susceptibility‐induced signal loss and improving the accuracy of single‐band and multi‐band fMRI. Numerical simulations, hybrid simulations, and in vivo fMRI experiments were conducted to compare conventional magnitude‐based super‐resolution MRI with a proposed complex‐valued super‐resolution approach. Simulations used a high‐resolution digital phantom and experimentally acquired gradient‐echo echo‐planar imaging (EPI) data at 3 T. Functional MRI experiments were performed in six healthy participants on a 3 T Siemens Skyra system with single‐band and multi‐band EPI acquisitions. Preprocessing included Nyquist artifact correction, motion correction, and B 0 drift correction. Complex‐valued data were preserved throughout the reconstruction, and coil sensitivity profiles were incorporated into the reconstruction pipelines. Simulations demonstrated that the complex‐valued reconstruction reduced susceptibility‐related signal loss compared with the magnitude‐based approach, particularly at longer echo times and with multi‐band EPI data. In vivo results confirmed these findings, showing improved preservation of high‐resolution features and reduced artifacts. Group‐level fMRI analyses revealed robust detection of motor‐related activation with complex‐valued super‐resolution methods, with higher statistical power observed in multi‐band acquisitions. In conclusion, the proposed complex‐valued super‐resolution MRI framework improves the quality and reliability of reconstructed fMRI data by reducing susceptibility artifacts and preserving fine‐scale spatial information. This approach has potential to enhance the spatial resolution of fMRI studies without substantially increasing scan time and can be extended to different acquisition schemes and sub‐voxel shifting strategies.