Noninvasive differentiation of benign and malignant solid pulmonary nodules using multiparameter dual-layer spectral CT radiomics

Abstract

To establish and validate a noninvasive multiparametric radiomics model based on dual-layer spectral CT (DLCT) for distinguishing preoperatively malignant from benign solid pulmonary nodules (SPNs). This retrospective study enrolled 441 patients with pathologically confirmed SPNs who underwent preoperative DLCT and were divided into training (n = 252), internal test (n = 112), and external test (n = 77) cohorts. Radiomics features were extracted from conventional and virtual monoenergetic images (40 and 70 keV) and material decomposition images (including iodine density (ID), Z-effective atomic number (Zeff), and electron density (ED) maps) in arterial (AP) and venous (VP) phases. Logistic regression was used to construct radiomics models, and a combined clinical-radiomics model was visualized as a nomogram. Subgroup analysis was performed by nodule size (≤ 10 mm vs.  > 10 mm), and diagnostic accuracy was compared with that of two radiologists. The optimal radiomics model, comprising features from ID in VP and Zeff in both AP and VP, achieved an area under the curve (AUC) of 0.835, 0.804, and 0.772 in the training, internal, and external cohorts, respectively. The combined model (age, lobulation, and ten radiomic features) outperformed the clinical-radiological model in all cohorts (AUC: 0.889 vs. 0.816; 0.865 vs. 0.795; 0.825 vs. 0.743; p < 0.05). It maintained strong performance for ≤ 10 mm and > 10 mm nodules, with AUCs of 0.875 and 0.888 (internal test) and 1.000 and 0.794 (external test). A DLCT-based multiparametric radiomics model, integrated with clinical-radiological features, enables accurate preoperative, noninvasive differentiation between benign and malignant SPNs, including subcentimeter nodules. Question: Accurate preoperative differentiation of solid pulmonary nodules remains clinically challenging because benign and malignant nodules often show overlapping features on conventional CT. Findings: A combined model integrating multiparameter dual-layer spectral CT radiomics and clinical-radiological features outperformed the clinical-radiological model in differentiating benign and malignant solid pulmonary nodules. Critical relevance statement: The developed model provides an accurate, noninvasive tool for preoperative differentiation of indeterminate solid pulmonary nodules, supporting clinical management decisions. Question: Accurate preoperative differentiation of solid pulmonary nodules remains clinically challenging because benign and malignant nodules often show overlapping features on conventional CT. Findings: A combined model integrating multiparameter dual-layer spectral CT radiomics and clinical-radiological features outperformed the clinical-radiological model in differentiating benign and malignant solid pulmonary nodules. Critical relevance statement: The developed model provides an accurate, noninvasive tool for preoperative differentiation of indeterminate solid pulmonary nodules, supporting clinical management decisions.