Prediction of pathological grading of high-risk solitary pulmonary nodules based on CT imaging deep learning: A multi-centric study.

Abstract

e13577 Background: Histopathological classification of operable early-stage lung adenocarcinoma may influence the intraoperative surgical decision and postoperative management for patients. However, it is difficult to comprehensively evaluate the grade of risk of operable lung adenocarcinoma based on preoperative CT imaging alone. Therefore, we aimed to establish a pathological grading model for patients with pulmonary nodules through preoperative CT imaging. Methods: This was a diagnostic, retrospective, multi-centric study conducted in two independent centers from January 1, 2019, to June 1, 2021. After inclusion and exclusion criteria, a total of 396 patients were included from two centers and were diagnosed as stage IA lung invasive adenocarcinoma (IAC). Patients (n=308) from the Guangdong Provincial People's Hospital with solitary pulmonary nodules were randomly divided into the training cohort (n=222) and the internal validation cohort (n=86). Patients from the Jiangxi Cancer Hospital comprised the external cohort (n=88). Their clinical characteristics and preoperative CT images were collected at the same high-quality standard. Extracted radiomic features from tumor and peritumor areas were selected via the least absolute shrinkage and selection operator (LASSO) algorithm for a radiomics model (RM) construction. Logistic Regression analysis was conducted to develop a novel multi-parameter model (MPM) that integrated clinical characteristics and radiomic features. The area under the curve (AUC) of the receiver operator characteristic curve (ROC) was calculated to compare the prediction performance of models. Results: The MPM composed of clinical characteristics and radiomic features performed better than RM. It achieved good predictive value for high-risk IAC. The AUC of MPM in the training cohort was 0.87 (95%CI, 0.81-0.92), which achieved better identification performance than the RM (See Table). Similarly, model comparisons were conducted in the two validation cohorts in which the MPM gained high AUC values of 0.78 (95%CI, 0.68-0.88) and 0.83 (95%CI, 0.74-0.91), respectively. In addition, the sensitivity of MPM was higher than 80.0% in the three cohorts. Conclusions: The MPM may be a practical and reliable tool for predicting the pathological risk of patients with solitary pulmonary nodules. [Table: see text]