AI-Accelerated 3D Gradient Echo versus Ultrashort Echo Time MRI for Lung Nodule Detection and Measurement

Background Ultra-short echo time (UTE) sequences are widely used in pulmonary MRI due to favorable lung parenchyma signal but are limited by acquisition time and complex reconstruction. Purpose To compare an AI-enhanced 3D gradient echo sequence with a UTE sequence for lung nodule detection and Lung-RADS classification, using CT as the reference standard. Methods In this single-center observational study, patients with at least one CT-detected lung nodule underwent pulmonary MRI. MRI included a respiratory-gated AI-accelerated 3D gradient-echo sequence (CSAI-GRE) and a respiratory-gated 3D ultrashort echo time sequence (UTE), both acquired with comparable scan times. Three radiologists independently assessed image quality, nodule detection, and Lung-RADS classification. CT served as the reference standard. Statistical analyses included paired tests for image quality, mixed-effects modeling for lesion-level detection, and κ statistics for Lung-RADS agreement. Results This study included 54 patients (mean age, 65 ± 12 years; 23 women [43%]) with 97 pulmonary nodules. Subjective image quality was higher for CSAI-GRE than UTE (mean score, 4.30 vs 3.91; p = .005). Lesion-level detection was 96.9% for CSAI-GRE and 92.8% for UTE; mixed-effects analysis showed higher conditional detection odds for CSAI-GRE (OR, 10.2; 95% CI, 2.4-43.3; p = .002). MRI–CT agreement for Lung-RADS categorization ranged from substantial to almost perfect across readers (κ range, 0.79–0.93 for CSAI-GRE; 0.72–0.94 for UTE). Lung-RADS reclassification relative to CT occurred in 12.3% (20/162; 95% CI, 8.1-18.3) of CSAI-GRE and 20.8% (33/159; 95% CI, 15.2–27.7) of UTE reader-assessments, including 5.6% (9/162; 95% CI, 3.0–10.2%) and 10.7% (17/159; 95% CI, 6.8–16.5%) management-relevant changes, respectively. Conclusion CSAI-GRE and UTE enabled reliable pulmonary nodule detection. CSAI-GRE provided superior subjective image quality and slightly higher detection performance than UTE. However, the Lung-RADS reclassification rates suggest that MRI-based Lung-RADS categorization may not yet be sufficiently robust for routine clinical application.