Predicting higher-risk growth patterns in invasive lung adenocarcinoma with multiphase multidetector computed tomography and 18F-fluorodeoxyglucose PET radiomics

Abstract

Purpose: To develop a predictive model for identifying the higher-risk growth pattern of invasive lung adenocarcinoma using multiphase multidetector computed tomography (MDCT) and 18F-fluorodeoxyglucose (FDG) PET radiomics.

Methods: A total of 203 patients with confirmed invasive lung adenocarcinoma between January 2018 and December 2021 were enrolled and randomly divided into training (n = 143) and testing sets (n = 60). Patients were classified into two groups according to the predominant growth pattern (lower-risk group: lepidic/acinar; higher-risk group: papillary/solid/micropapillary). Preoperative multiphase MDCT and 18F-FDG PET images were evaluated. The Artificial Intelligence Kit software was used to extract radiomic features. Five predictive models [arterial phase, venous phase, and plain scan (AVP), PET, AVP-PET, clinical, and radiomic-clinical (Rad-Clin) combined model] were developed. The models' performance was assessed using receiver-operating characteristic (ROC) curves and compared using the DeLong test.

Results: Among the radiomics models (AVP, PET, and AVP-PET), the AVP-PET model [area under ROC curve (AUC) = 0.888] outperformed the PET model (AUC = 0.814; P = 0.015) in predicting the higher-risk growth patterns. The combined Rad-Clin model (AUC = 0.923), which integrates AVP-PET radiomics and five independent clinical predictors (gender, spiculation, long-axis diameter, maximum standardized uptake value, and average standardized uptake value), exhibited superior performance in predicting the higher-risk growth pattern compared with radiomic models (P = 0.043, vs. AVP-PET; P = 0.016, vs. AVP; P = 0.002, vs. PET) or the clinical model alone (constructing based on five clinical predictors; AUC = 0.793; P < 0.001).

Conclusion: The combined Rad-Clin model can predict the higher-risk growth patterns of invasive adenocarcinoma (IAC). This approach could help determine individual therapeutic strategies for IAC patients by distinguishing predominant growth patterns with high risk.