Deep learning-based super-resolution ultrashort echo time MRI combined with free-breathing Star-VIBE dynamic contrast-enhanced MRI for characterization of Lung-RADS 4 pulmonary nodules.
Authors
Affiliations (1)
Affiliations (1)
- Department of Radiology, The Fourth Hospital of Hebei Medical University, Shijiazhuang, China.
Abstract
Lung Imaging Reporting and Data System (Lung-RADS) 4 pulmonary nodules require further evaluation because of indeterminate malignancy risk. However, repeated computed tomography (CT) examinations raise concerns about cumulative radiation exposure, and ultrashort echo time magnetic resonance imaging (UTE-MRI) remains limited by spatial resolution and motion sensitivity. This study aimed to assess the diagnostic performance of generative adversarial network (GAN)-based super-resolution UTE MRI (SR-UTE-MRI) in combination with star-volumetric interpolated breath hold examination (Star-VIBE) dynamic contrast-enhanced MRI (DCE-MRI) for such nodules, using CT and histopathological findings as reference standards. In this prospective study, patients with Lung-RADS 4 pulmonary nodules underwent CT, free-breathing UTE-MRI with subsequent SR reconstruction, and Star-VIBE DCE-MRI. Nodule size, morphological features, and perfusion parameters were assessed. Diagnostic performance of individual imaging metrics and a combined model was evaluated using receiver operating characteristic (ROC) curve analysis. A total of 91 patients (43 men and 48 women; mean age, 56.2±13.9 years) with 91 nodules (27 part-solid and 64 solid; mean diameter, 24.1±8.9 mm) were finally included. Measurements derived from UTE-MRI underestimated mean nodule diameter relative to CT, resulting in misclassification in a subset of nodules. SR-UTE-MRI images reduced measurement bias and improved Lung-RADS classification accuracy and visualization of morphological features. Malignant pulmonary nodules demonstrated significantly higher volume transfer constant (K<sup>trans</sup>) and rate constant (K<sub>ep</sub>) values compared with benign nodules (K<sup>trans</sup>: 0.28 ± 0.19 min<sup>-1</sup> <i>vs.</i> 0.16±0.11 min<sup>-1</sup>, P<0.001; K<sub>ep</sub>: 0.79±0.38 min<sup>-1</sup> <i>vs.</i> 0.53±0.23 min<sup>-1</sup>,P<0.001). Among individual parameters, K<sup>trans</sup> exhibited the highest diagnostic performance [area under the curve (AUC) =0.799; cut-off, 0.14 min<sup>-1</sup>]. A multiparametric model incorporating perfusion parameters, nodule size, and spiculation yielded superior discriminatory performance (AUC =0.946; sensitivity =91%; specificity =88%; accuracy =90%). GAN-based SR-UTE reduced the underestimation of lesion size associated with UTE-MRI and improved the accuracy of Lung-RADS classification. Star-VIBE DCE-MRI provided quantitative functional biomarkers for the assessment of malignancy. The integration of structural and functional imaging parameters yielded superior diagnostic performance compared with single-modality or single-parameter approaches, highlighting its potential as a comprehensive, non-ionizing imaging strategy for evaluation of indeterminate pulmonary nodules.