Room: Exhibit Hall | Forum 6
Purpose: To investigate the feasibility of using quantitative radiomic features and clinical variables to differentiate non-small cell lung cancer (NSCLC) subtypes.
Methods: One hundred patients with histologically confirmed NSCLC were retrospectively evaluated (60 adenocarcinoma (ADC) and 40 squamous cell carcinoma (SqCC) patients). Gross tumor volume (GTV) was delineated on positron emission tomography (PET) images. In total, 107 features were extracted, including 60 texture and 47 metabolic features. The least absolute shrinkage and selection operator (LASSO) method was used to select the optimal feature set. We analyzed the differences between these features in the two tumor types. Multivariable logistic regression analysis was performed using the radiomic signature and clinical variables to develop the differentiation model. The modelâ€™s discriminatory performance was assessed by analyzing the area under the receiver operating characteristic (ROC) curve (AUC).
Results: In total, 107 features were extracted, and five features were selected as the best discriminators for the radiomic model, including four texture features and one metabolic feature. The five characteristics were all significantly different between ADC and SqCC subgroups. The rate of change of ADC features relative to SqCC features ranged from -38.73% to 47.74%, and high-intensity short-run and entropy showed the highest rates of change. The radiomic signature performance yielded AUCs of 0.804 and 0.700 in the training and validation datasets, respectively. When clinical factors were integrated, AUC values in the training and validation groups reached 0.822 and 0.781, respectively.
Conclusion: The 18F-FDG PET radiomics predictive model is a promising and applicable adjunct pretreatment approach to identify ADC and SqCC subtypes of NSCLC.
PET, Texture Analysis, Quantitative Imaging