Evaluation of a dual-energy computed tomography parameter -radiomics combined model for solitary pulmonary nodules.
To construct a dual-energy computed tomography (DECT) parameter-radiomics combined model based on DECT quantitative parameters and radiomic features and to evaluate its diagnostic performance in differentiating benign from malignant solitary pulmonary nodules.
We retrospectively collected data from inpatients who underwent contrast-enhanced thoracic DECT scans at our hospital between June 2018 and November 2024. Quantitative parameters measured and calculated included iodine concentration, effective atomic number, normalized iodine concentration, normalized effective atomic number, spectral Hounsfield unit slope, and extracellular volume fraction. These DECT parameters were used to construct a predictive model. The dataset was randomly divided into training and test sets in a 7:3 ratio. Radiomic features were extracted from mixed-energy images of the arterial and venous phases (A120kVp and V120kVp) and from 70-keV monoenergetic images (A70keV and V70keV). Six machine-learning algorithms were used to construct radiomics models, and the optimal model was selected based on performance. This optimal radiomics model was then combined with the DECT parameter model to create the combined model. Model performance was evaluated using accuracy, sensitivity, specificity, and area under the curve (AUC), and model comparisons were performed using the DeLong test (Python). A P-value of<0.05 was considered statistically significant.
In total, 209 patients were enrolled (166 in the malignant group and 43 in the benign group). In the test set, the diagnostic performance of the combined model (AUC = 0.906) exceeded that of the DECT-only radiomics model (AUC = 0.794). The combined model demonstrated an accuracy of 82.5%, sensitivity of 80.0%, and specificity of 92.3%.
The DECT parameter-radiomics combined model effectively differentiated benign from malignant solitary pulmonary nodules and shows promising potential for clinical translation.
We retrospectively collected data from inpatients who underwent contrast-enhanced thoracic DECT scans at our hospital between June 2018 and November 2024. Quantitative parameters measured and calculated included iodine concentration, effective atomic number, normalized iodine concentration, normalized effective atomic number, spectral Hounsfield unit slope, and extracellular volume fraction. These DECT parameters were used to construct a predictive model. The dataset was randomly divided into training and test sets in a 7:3 ratio. Radiomic features were extracted from mixed-energy images of the arterial and venous phases (A120kVp and V120kVp) and from 70-keV monoenergetic images (A70keV and V70keV). Six machine-learning algorithms were used to construct radiomics models, and the optimal model was selected based on performance. This optimal radiomics model was then combined with the DECT parameter model to create the combined model. Model performance was evaluated using accuracy, sensitivity, specificity, and area under the curve (AUC), and model comparisons were performed using the DeLong test (Python). A P-value of<0.05 was considered statistically significant.
In total, 209 patients were enrolled (166 in the malignant group and 43 in the benign group). In the test set, the diagnostic performance of the combined model (AUC = 0.906) exceeded that of the DECT-only radiomics model (AUC = 0.794). The combined model demonstrated an accuracy of 82.5%, sensitivity of 80.0%, and specificity of 92.3%.
The DECT parameter-radiomics combined model effectively differentiated benign from malignant solitary pulmonary nodules and shows promising potential for clinical translation.