Development and validation of a multimodal predictive model based on clinical, biochemical, and quantitative...

This study aimed to develop and validate a clinical prediction model integrating clinical characteristics, biochemical markers, and quantitative dual-energy CT (DECT) parameters to differentiate malignant from benign thyroid nodules.

This retrospective study included 172 patients with thyroid nodules (87 malignant and 85 benign). All patients underwent non-contrast and dual-phase contrast-enhanced dual-energy CT (DECT) of the thyroid. Candidate features for model development included clinical variables (sex and age), biochemical markers (CEA, TPOAb, FT4, TSH, FT3, TRAb, Tg, CT, and TgAb), and spectral CT-derived quantitative parameters (thyroid nodule volume [TV], spectral curve slope, effective atomic number, calcium, hydroxyapatite [HAP], iodine concentration, and ICDNR). The patients were randomly divided into a training cohort and a validation cohort at a 7:3 ratio. Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) and the Boruta algorithm. A predictive model was then developed and internally validated. Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA) to assess discrimination, calibration, and clinical utility.

Multivariable logistic regression identified age (OR = 0.93, p < 0.001), TSH (OR = 1.65, p = 0.037), and thyroid nodule volume (TV, OR = 0.91, p = 0.017) as independent predictors of malignancy. HAP-MN and ICDNR were retained as auxiliary predictors following LASSO/Boruta feature selection and sensitivity analysis. The final model demonstrated good discriminative performance, with an AUC of 0.866 (95% CI: 0.802-0.930) in the training cohort and 0.852 (95% CI: 0.742-0.961) in the validation cohort. Model diagnostics indicated an acceptable fit. Goodness-of-fit tests and calibration curves showed good agreement between predicted and observed outcomes, and DCA further confirmed the model's favorable clinical utility.

In this study, we successfully developed a multimodal predictive model integrating clinical, biochemical, and spectral CT-derived quantitative features, which demonstrated excellent diagnostic accuracy for thyroid nodules. This non-invasive and objective tool may improve risk stratification, reduce unnecessary interventions, and support personalized patient management.