An AI-enhanced lung MRI sequence achieved higher image quality and slightly better pulmonary nodule detection than ultrashort echo time MRI. A single-centre study from Technical University of Munich, published in Radiology Advances, compared both respiratory-gated techniques with chest CT. Agreement was strong for nodule measurement and Lung-RADS categorisation, but management-relevant reclassifications occurred with both sequences. Further optimisation is therefore needed before MRI-based assessment can support routine clinical decisions.
Comparable MRI Sequences Tested Against CT
Pulmonary MRI has gained clinical utility as imaging techniques have improved. Ultrashort echo time MRI is widely used because it can visualise lung parenchyma signal, but it relies on non-Cartesian sampling and complex reconstruction. The AI-enhanced gradient echo approach uses Cartesian sampling with a reconstruction framework combining parallel imaging, compressed sensing and deep learning. The comparison therefore focused on whether a highly accelerated gradient echo sequence could match or exceed ultrashort echo time MRI for nodule detection, measurement and Lung-RADS classification under similar respiratory gating and scan duration. The gradient echo sequence used 1 mm isotropic voxels, while the ultrashort echo time sequence used lower isotropic resolution to maintain comparable acquisition time.
Must Read: Explainable AI Supports Lung Cancer CT Detection
The Munich cohort included adults with at least one pulmonary nodule detected on clinically indicated chest CT who then underwent both MRI sequences within 8 days. Nodules larger than 3 mm on CT were included, with no maximum size criterion. Patients with advanced chronic lung conditions likely to impair nodule detection, acute respiratory infections, pleural effusion, pregnancy or standard MRI contraindications were excluded. The final cohort comprised 54 patients with 97 pulmonary nodules. Three radiologists independently assessed the MRI datasets, separated by a washout period, while a cardiothoracic radiologist assessed CT without access to MRI findings. Readers evaluated image quality, nodule detection, attenuation, mean axial diameter and Lung-RADS categories.
AI-Enhanced Sequence Shows Higher Detection
The AI-enhanced gradient echo sequence achieved higher subjective image quality than ultrashort echo time MRI across all assessed domains, including interpretability, artefacts, blurring, anatomical detail and nodule conspicuity. Mean image quality scores were higher for the gradient echo sequence, with both MRI techniques remaining within diagnostically sufficient ranges under the five-point scoring system. The difference was consistent with the technical design of the gradient echo acquisition, which used smaller isotropic voxels while maintaining a scan time similar to the ultrashort echo time sequence.
Detection performance was high for both approaches, but the AI-enhanced sequence detected more nodules than ultrashort echo time MRI. Across reader assessments, detection was 96.9% for the gradient echo sequence and 92.8% for ultrashort echo time MRI. The difference remained evident in nodules smaller than 10 mm, where detection reached 94.4% and 87.0%, respectively. No false-positive nodules were recorded for either sequence. Missed lesions were small solid nodules, and no part-solid or ground-glass nodules were missed.
Agreement with CT for nodule diameter was excellent for both MRI sequences. Reader-averaged MRI measurements showed no systematic measurement bias against CT, although variability remained relevant, especially for small nodules. Agreement between MRI and CT for nodule attenuation was stronger for the gradient echo sequence than for ultrashort echo time MRI. Inter-reader agreement for attenuation was nearly perfect for both techniques.
Classification Changes Limit Clinical Use
Lung-RADS categorisation showed substantial to almost perfect agreement between MRI and CT for both sequences. Inter-reader agreement was almost perfect for the AI-enhanced gradient echo sequence and substantial for ultrashort echo time MRI. Even so, reclassification against CT occurred in a minority of assessments and was more frequent with ultrashort echo time MRI. Category changes occurred in 12.3% of gradient echo assessments and 20.8% of ultrashort echo time assessments. Management-relevant changes occurred in 5.6% and 10.7%, respectively.
Most reclassifications involved downgrades from Lung-RADS 4X to lower categories or upgrades from 4A to 4B. These changes matter because the classification framework links categories to clinical management. Reliable MRI-based detection was supported, but routine reliance on MRI for management categorisation was not. Lung-RADS was used because it offers a granular framework that parallels other pulmonary nodule guidance, including Fleischner Society and British Thoracic Society recommendations.
Potential applications remain focused on contexts where MRI is already being performed or where repeated imaging is a concern. Lung assessment during abdominal or pelvic MRI for oncologic staging or follow-up could reduce the need for additional CT imaging. MRI may also have a role in paediatric populations or patients requiring repeated examinations. However, several limitations restrict generalisability. The cohort was enriched for CT-detected nodules, advanced emphysema and severe structural lung disease were excluded, specificity could not be assessed, and histology was not available for all nodules.
AI-enhanced 3D gradient echo MRI provides high image quality and reliable pulmonary nodule detection, with measurement agreement against CT and slightly better detection than ultrashort echo time MRI in this single-centre cohort. The results support further evaluation of accelerated lung MRI, particularly in oncologic follow-up and settings where repeated imaging is relevant. Larger and more heterogeneous cohorts are needed to test performance in screening and follow-up populations. However, management-relevant Lung-RADS reclassifications and measurement variability show that MRI-based nodule assessment needs further optimisation before routine clinical decision-making can depend on it.
Source: Radiology Advances
Image Credit: iStock
References:
Marka AW, Weiss K, Rosenkranz H et al. (2026) AI-Accelerated 3D Gradient Echo versus Ultrashort Echo Time MRI for Lung Nodule Detection and Measurement. Radiology Advances: umag029.
