Deep learning reconstruction improves detection of focal liver lesions in hepatobiliary phase compared to conventional EOB-MRI.

Objective: In recent years, the application of Deep Learning Reconstruction (DL Recon) technology in MRI has significantly improved the detection rate of small focal liver lesions (FLLs). This study aims to compare the detection performance of DL Recon EOB-MRI with traditional Gd-EOB-DTPA-enhanced liver MRI during the hepatobiliary phase (HBP) for FLL detection. Methods: This prospective, single-center study included 53 patients who underwent EOB-MRI during the hepatobiliary phase (HBP). For each patient, four types of images were acquired: 3 mm thickness non-DL reconstruction (Standard Non-DL), 3 mm thickness DL reconstruction (Standard DL), 1 mm thickness non-DL reconstruction (HR Non-DL), and 1 mm thickness DL reconstruction (HR-DL). Lesions were categorized based on signal intensity into high-signal and low-signal types, and classified into five size groups: >30 mm, 20-30 mm, 10-20mm, 5-10mm, and <5 mm. Three abdominal radiologists independently assessed the number of FLLs on HBP images. Pairwise comparisons between the four types of images were made using the Wilcoxon signed-rank test and Generalized Estimating Equations to assess differences. Results: The intraclass correlation coefficient (ICC) between all groups was >0.90, indicating excellent consistency ( Conclusion: DL-Recon combined with MR techniques demonstrates significant clinical value in enhancing the detection of small low-signal lesions in EOB-MRI hepatobiliary phase, significantly improving the detection rate of FLLs.