CT-based prediction models improve malignancy prediction in surgically resected part-solid pulmonary nodules

ObjectiveTo develop and validate computed tomography (CT)-based prediction models for malignancy in part-solid pulmonary nodules (PSNs).MethodsIn this retrospective study, 204 surgically resected PSNs (107 malignant, 97 benign) were analyzed. Clinical data and CT morphological features were evaluated. Vascular patterns were classified into five types (I-V). Quantitative vascular counts (N1-N5, TN) were recorded. Nodules were randomly split into training (n = 143) and testing (n = 61) cohorts. Three logistic regression models were constructed: Model 1 (baseline clinical and morphological features), Model 2 (Model 1 + qualitative vascular types IV and V), and Model 3 (Model 1 + quantitative vessel counts N1-N3). Model performance was assessed using the area under the receiver operating characteristic curve (AUC), calibration (Hosmer-Lemeshow test), and clinical utility (decision curve analysis).ResultsMalignant nodules were associated with older age (59 ± 10 vs. 56 ± 11 years, p = 0.018), female predominance (62.6% vs. 43.3%, P = 0.006), and specific CT features including irregular shape, lobulation, spiculation, vacuole sign, and pleural indentation (all P < 0.05). Vascular patterns IV (interruption) and V (distortion) were significantly more prevalent in malignant nodules (43.9% vs. 14.4%, and 51.4% vs. 5.2%, respectively; both P < 0.001). Quantitative counts of interrupted (N4) and distorted (N5) vessels were also significantly higher in malignancies (P < 0.001). In multivariable analysis, Model 2, incorporating vascular types IV and V, demonstrated superior predictive performance with a training AUC of 0.916 (95% CI: 0.872–0.960) and a testing AUC of 0.898 (95% CI: 0.821–0.974), significantly outperforming Model 1 (AUC 0.860/0.827) and Model 3 (AUC 0.866/0.823) (DeLong test, P = 0.012 and P = 0.040). Model 2 also showed excellent calibration and provided the highest net clinical benefit across a wide range of threshold probabilities.ConclusionQualitative CT assessment of vascular interruption and distortion (types IV and V) significantly improves the prediction of malignancy in PSNs over conventional morphological features alone. A model integrating these vascular patterns offers excellent diagnostic accuracy and clinical utility, potentially aiding in the preoperative risk stratification of PSNs.