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

• DOI: 10.1200/jco.2022.40.16_suppl.e13577
• Publication Date: 2022-06-06T15:57:32Z
• Authors: Weihuan Lin; Weijie Zhan; HaiYu Zhou
• Keywords: Grading (engineering)

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