Dual-energy CT Radiomics Combined with Quantitative Parameters for Differentiating Lung Adenocarcinoma From Squamous Cell Carcinoma: A Dual-center Study
Ze Lin,Ying Liu,Chengcheng Xia,Pei Huang,Zhiwei Peng,Li Yi,Yu Wang,Xiao Yu,Bing Fan,Minjing Zuo
DOI: https://doi.org/10.1016/j.acra.2024.09.024
2024-09-25
Abstract:Rationale and objectives: To evaluate the ability of dual-energy CT(DECT)-based quantitative parameters and radiomics features to differentiate solid lung adenocarcinoma (ADC) from squamous cell carcinoma (SCC). Methods: This study included 213 patients diagnosed with ADC and SCC who underwent DECT scans at two centers from November 2022 to December 2023. Patients at center 1 were randomly divided into training (n = 114) and internal test set (n = 50) in a 7:3 ratio, with center 2 serving as the external test set (n = 49). Radiologic and clinical data were combined to establish a clinical-radiologic model. Ten types of DECT energy images including conventional images, iodine density (ID), effective atomic number (Zeff), electron density, and virtual mono-energetic images (VMI) were reconstructed in both arterial phases (AP) and venous phases (VP). Quantitative parameters were measured at the uniform enhanced solid portion of the tumor and normalized to the aorta, used to develop a quantification model and calculate the quantitative score (quantscore). Radiologists manually delineated the tumor ROI at the largest level for extracting radiomics features in these 10 energy images. These features were used to establish 10 uni-energy models from which the best-performing features were selected to construct the final radiomics model and calculate a radiomics score (radscore). Then, a combined model was developed using the akaike information criterion(AIC) and compared to the clinical-radiological model to test its diagnostic validity. Results: The independent predictors of the clinical-radiological model included age, gender, and central or peripheral location, and the AUCs for the training set, internal test set, and external test set were 0.808, 0.837, and 0.802. The quantification model incorporated 40 keV CT values, Zeff, normalized Zeff, and the slope of the spectral attenuation curve (λHU) in the AP and normalized ID, Zeff, and λHU in the VP. Uni-energy models based on AP ID maps, AP Zeff maps, and VP VMI 65 keV significantly outperformed AUC= 0.5, and 11 radiomics features were selected from these three models to construct the final radiomics model. The combined model, incorporating age, gender, quantscore, and radscore, significantly outperformed the clinical-radiological model in the training set (AUC=0.952 vs 0.808, P < 0.001), and demonstrated higher performance in both the internal and external test sets, although these differences did not reach statistical significance (AUC=0.870 vs 0.837, for the internal test set [P = 0.542], 0.888 vs 0.802 for the external test sets [P = 0.128]). The evaluation of the combined model demonstrated good discriminative ability and potential for generalization. Conclusion: The combined model, integrating quantitative parameters and radiomics features from DECT multi-energy images with clinical-radiological characteristics, can be used as a non-invasive tool to differentiate ADC from SCC.