CT-based Radiologic Ternary Classification Model in Predicting Pathologic Invasiveness of Pulmonary Nonsolid Nodules

Background Evaluating the extent of invasiveness for nonsolid nodules (NSNs) in patients with lung adenocarcinoma at CT could affect clinical decision-making but can be challenging. Purpose To investigate CT characteristics of NSNs associated with pathologic invasiveness and to develop a radiologic ternary classification model for differentiating among preinvasive lesions, minimally invasive adenocarcinoma (MIA), and invasive adenocarcinoma (IAC). Materials and Methods This retrospective study enrolled patients with pathologically confirmed lung adenocarcinoma and suspicious malignant NSNs measuring 3.0-30.0 mm on preoperative CT scans between January 2012 and June 2024. For each NSN, the size, location, margin, shape, nodule CT attenuation, uniformity of density, lobulation sign, reticulation sign, intranodular vessels, bubble-like lucency sign, air bronchogram sign, and pleural retraction sign were independently evaluated by two radiologists blinded to clinical information and pathology results. Univariable ordinal regression and partial proportional odds model analyses were performed. Three nested mixed-effects models were compared in differentiating pathologic invasiveness subtypes. Results This study included 1683 patients (median age, 53 years [IQR, 45-61 years]; 1145 women) with 2125 NSNs. Partial proportional odds model analysis demonstrated that the independent radiologic factors for predicting pathologic invasiveness were average diameter (preinvasive lesion vs MIA: odds ratio [OR], 1.34; MIA vs IAC: OR, 1.54), intranodular vessels (one vessel: OR, 2.22; two vessels: OR, 3.06; more than two vessels: OR, 25.16), mean CT attenuation (OR, 1.54), heterogeneous density (OR, 2.45), spiculation (OR, 1.72), lobulation (OR, 1.50), pleural retraction (OR, 1.43), bubble lucency (OR, 1.81), and air bronchogram (OR, 1.74). The overall diagnostic performance of the radiologic ternary classification model was excellent (C index, 0.92; 95% CI: 0.91, 0.92). Incorporating mean CT attenuation and morphologic features improved model performance in predicting NSN pathologic invasiveness compared with using nodule diameter alone (all <i>P</i> < .001). Conclusion The radiologic ternary classification model demonstrated excellent diagnostic performance in differentiating among preinvasive lesions, MIA, and IAC in NSNs detected on CT images. © RSNA, 2025 <i>Supplemental material is available for this article.</i> See also the editorial by Arita and Schalekamp in this issue.