Optimized dual-source dual-energy computed tomography nomogram model integrating background normalization improves solid thyroid nodules diagnosis.

Background: Accurate preoperative diagnosis of benign and malignant thyroid nodules is crucial for personalized patient treatment and management. This study aimed to create a dual-source dual-energy computed tomography (DS-DECT) based nomogram model to predict the risk of malignant thyroid nodules. Methods: A total of 263 patients (288 nodules) with thyroid nodules who underwent preoperative neck DS-DECT scans and were pathologically confirmed were included in this study. The computed tomography (CT) radiological features and DS-DECT-derived quantitative parameters of the nodules were collected. Subsequently, the thyroid nodules were randomly partitioned into a training cohort (n=201) and a validation cohort (n=87) at a ratio of 7:3. Univariate logistic regression analysis identified predictors (P<0.05), followed by least absolute shrinkage and selection operator (LASSO) logistic regression to screen features in the training cohort. Multivariate logistic regression analysis was then conducted to determine independent predictors of malignancy (P<0.05) and to construct a nomogram model for predicting malignancy risk. The performance of the model was assessed using the area under the curve (AUC), calibration curve, and decision curve analysis (DCA). The applicability of the nomogram was evaluated through internal validation. Results: In the final stepwise multivariable logistic regression model, independent predictors of malignancy included iodine concentration in the arterial phase [IC_IAP; odds ratio (OR) =0.286; 95% confidence interval (CI): 0.140-0.517; P<0.001], normalized iodine concentration relative to the thyroid parenchyma in the arterial phase (NIC_P_IAP; OR =0.133; 95% CI: 0.033-0.417; P=0.003), effective atomic number in the venous phase (Zeff_IVP; OR =0.137; 95% CI: 0.050-0.333; P<0.001), thyroid edge interruption (OR =3.791; 95% CI: 1.599-9.491; P=0.003), and enhanced blurring (OR =3.247; 95% CI: 1.373-7.937; P=0.008). The AUCs of the nomogram model, based on these five factors, were 0.932 (95% CI: 0.898-0.963) for the training set and 0.908 (95% CI: 0.842-0.964) for the validation set. The Hosmer-Lemeshow test indicated that the nomogram model had a good fit (P>0.05), and the calibration curve was close to the standard curve. DCA showed significant net benefits from using the model. Conclusions: The nomogram model, based on normalized multiphase quantitative DECT parameters and qualitative imaging features, serves as an effective diagnostic tool for imaging physicians to distinguish between benign and malignant nodules.