Dual-Layer Spectral CT-Based Models for the Differential Diagnosis of Breast Lesions

To develop a hybrid model integrating morphological features with quantitative parameters from dual-layer spectral computed tomography (DSCT) to noninvasively differentiate benign breast lesions from breast cancer.

This retrospective study included patients with breast lesions incidentally detected on spectral CT between January 2020 and May 2024. Lesions were categorized as benign or malignant based on histopathology or greater than two years' stability on follow-up imaging. Clinical variables, morphological characteristics, and DSCT parameters were collected. The cohort was randomly divided into training, validation, and independent testing sets using stratified sampling. Feature selection was performed exclusively within the training set using univariate analysis, collinearity assessment, and least absolute shrinkage and selection operator regression with the one-standard-error criterion. Clinico-radiological, DSCT-based, and hybrid models were constructed using multivariate logistic regression. Model discrimination was evaluated using receiver operating characteristic analysis and, model stability was assessed using nested cross-validation. A nomogram was constructed based on the best-performing model.

A total of 102 patients (mean age 52 ± 11 years) were included, including 27 patients (26%) with benign breast lesions and 75 patients (74%) with breast cancer. In the independent testing set, the DSCT-based model outperformed the clinical-radiological model (AUC = 0.967 vs 0.933), while the hybrid model achieved the highest diagnostic performance (AUC, 0.983). Calibration curves demonstrated strong concordance between predicted and observed outcomes. Decision curve analysis indicated favorable net benefit.

The hybrid model demonstrated good diagnostic performance in noninvasively differentiating benign from malignant breast lesions detected preoperatively on DSCT.