Our two-center study demonstrated that theplain &contrast-enhanced CTbased model with twenty-five CT variables which include four contrast-enhanced variables (enhanced CT value, enhancement rate, uniform enhancement, heterogeneous enhancement) had better prediction performance (0.88[95%CI, 0.82–0.93]) than only plain CT based model (0.93 [95%CI, 0.88–0.98]) for solitary solid pulmonary nodules.
Various models, including the Mayo Clinic model, the Veterans Affairs (VA) model, and the Brock model (PanCan model), have been developed utilizing clinical and CT characteristics to assess the malignancy of lung nodules20–22. The Mayo Clinic model identified age, smoking history, cancer history, nodule diameter, spiculation, and upper lobe as predictors of malignant nodules 20. The Brock model was developed to detect malignancy in nodules through low-dose CT screenings, utilizing predictors such as age, sex, family history of lung cancer, nodule location, emphysema, nodule size, and spiculation 21. The Veterans Affairs utilized logistic regression to design a model specifically for solitary nodules, estimating the likelihood of malignancy based on factors such as age, nodule diameter, smoking history, and time since quitting smoking22. However, prior research has demonstrated that while these models exhibit strong performance on their respective datasets, their utility for detecting large lung nodules is limited, necessitating optimization of model characteristics prior to clinical application23–27. Our study showed similar relative variables for predicting nodule malignancy in the model, like nodule location, nodule diameter, shape, age and gender.Furthermore, our model incorporates a greater number of semantic features, such as air bronchogram, pleural indentation, vascular invasion, postobstructive pneumonia, cavitation, necrosis, calcification, satellite nodules, and fat, as well as enhancement characteristics such as enhanced CT value, enhancement rate, uniform enhancement, and heterogeneous enhancement. The important of semantic features already be proved by previous study 28. Xiang et al. showed six radiological characteristics (diameter, lobulation, calcification, spiculation, pleural indentation, vascular invasion) were adopted as important predictors in their SVM model for the diagnosis of solid solitary pulmonary nodules with AUC 0.89. Our plain &contrast enhanced CT based model showed higher AUC 0.93, since we include more semantic features and enhancement characteristics (enhanced CT value, enhancement rate, uniform enhancement, heterogeneous enhancement).
The significance of CT enhancement level in the determination of malignancy in lung solid nodules has been established18,29. A lack of significant enhancement on contrast-enhanced CT (< 15HU) is indicative of a benign nodule. Consequently, contrast-enhanced CT has been widely utilized as the primary imaging examination technique prior to surgery, particularly in less developed nations12. Our study showed logistic regression model based on plain CT (no contrast enhanced CT) for predicting malignancy of solitary solid pulmonary nodules with sensitivity 0.85, specificity 0.84 and diagnostic accuracy 0.84 in development cohort and 0.79, 0.78, 0.79 inexternal validation cohort. When we added contrast-enhanced CT features into the model, it improves the diagnosis performance with sensitivity 0.91, specificity 0.87 and diagnostic accuracy 0.88 in development cohort and 0.88, 0.91, 0.90 inexternal validation cohort. This again suggests enhanced CT could be the basis for solitary solid pulmonary nodules preoperative diagnosis especially when preoperative biopsy and PET-CT are not applicable.
The present study has identified certain limitations. Firstly, the plain & contrast-enhanced CT model utilized in the study comprises only fundamental clinical information such as age and gender, while other factors such as smoking history, cancer history, and family history of cancer are worth considering for inclusion. Secondly, the study data solely comprised clinical patients, and thus, the efficacy of the model for lung cancer screening patients requires further verification. Additionally, the evaluation of the model was restricted to two datasets, and therefore, additional validation at various centers is necessary before its clinical application.
To conclude, a logistic regression model was constructed utilizing plain& contrast-enhanced CT characteristics, exhibiting superior efficacy in the assessment of malignancy in solitary solid lung nodules when compared to an only plain CT-based model. The utilization of this plain & contrast-enhanced CT model enables radiologists to provide recommendations concerning follow-up or surgical intervention for preoperative patient presenting with solid lung nodules.