Multi-Volume Isotropic Super-Resolution Wrist MRI Using Fourier-Guided Implicit Fusion

The human wrist comprises a complex network of fine structural components such as the Triangular Fibrocartilage Complex (TFCC). Comprehensive observation of this fine structure using magnetic resonance imaging (MRI) typically requires multiple orthogonal volumes of anisotropic resolutions. However, existing methods of implicit neural representation (INR)-based isotropic super-resolution often produce blurry results when handling multiple volume inputs due to their averaging effects. In this study, we propose a Fourier-guided implicit fusion method to enable INR-based methods to achieve high-quality reconstruction. By leveraging a 2D Fourier loss function, our approach enables the INR framework to integrate high-resolution signals from the lateral slices of low-resolution volumes, mitigating the blurring effects in the perpendicular direction while avoiding the averaging effect between multiple views. Our method achieves isotropic reconstruction for real wrist MRI volumes and facilitates downstream assessment of TFCC. Our method also outperforms competing methods on the HCP-1200 brain MRI dataset, demonstrating its generalization capability.