Learning Two-factor Representation for Magnetic Resonance Image Super-resolution

• DOI: 10.48550/arxiv.2409.09731
• Publication Date: 2024-09-15
• Authors: Weifeng Wei; Heng Chen; Pengxiang Su
• Keywords: Factor (programming language); Representation

Magnetic Resonance Imaging (MRI) requires a trade-off between resolution, signal-to-noise ratio, and scan time, making high-resolution (HR) acquisition challenging. Therefore, super-resolution for MR image is feasible solution. However, most existing methods face challenges in accurately learning continuous volumetric representation from low-resolution or require HR supervision. To solve these challenges, we propose novel method based on two-factor representation. Specifically, factorize intensity signals into linear combination of learnable basis coefficient factors, enabling efficient image. Besides, introduce coordinate-based encoding to capture structural relationships sparse voxels, facilitating smooth completion unobserved regions. Experiments BraTS 2019 MSSEG 2016 datasets demonstrate that our achieves state-of-the-art performance, providing superior visual fidelity robustness, particularly large up-sampling scale super-resolution.