Deep Learning-Reconstructed Reduced-field-of-view DWI in Rectal Cancer：Enhanced Rectal Wall Layering and Improved T Staging Accuracy

Motivation: Early-stage rectal cancer staging remains challenging with conventional T2-weighted imaging (T2WI), highlighting the need for high-quality diffusion-weighted imaging (DWI) to improve diagnostic accuracy. Goal(s): To assess the performance of deep learning reconstruction (DLR) combined with reduced-field-of-view (rFOV) in achieving high-quality, high-spatial-resolution DWI for rectal cancer. Approach: This study prospectively compared three DWI sequences: DLR-based rFOV (rFOVDL), standard rFOV (rFOVSTA), and standard full-field-of-view (fFOVSTA) DWI, assessing the image quality and regional staging performance. Results: rFOVDL DWI significantly improved spatial resolution and image quality, particularly enhancing the discriminability of mucosa-submucosa-muscularis layers, which in turn significantly improved T-staging accuracy, especially for early-stage tumors. Impact: This study demonstrated the clinical feasibility of DLR-enhanced rFOV DWI for rectal cancer. It improved the spatial resolution and the discriminability of mucosa-submucosa-muscularis layers, which facilitated higher T-staging accuracy, especially in early-stage tumors.