CT-based habitat radiomics for differentiating papillary thyroid carcinoma from nodular goiter: a two-center study

Abstract

Rationale and objectives

To develop habitat-based radiomics signatures for distinguishing papillary thyroid carcinoma (PTC) from nodular goiter (NG).

Material and methods

A retrospective study was conducted on PTC and NG patients from two centers. Univariable and multivariable logistic regression analyses were performed to identify independent risk factors for developing the clinical model. Tumor and ablation regions of interest (ROI) were split into three spatial habitats through K-means clustering algorithm and dilated with 2 mm, 4 mm 6 mm, and 8 mm thicknesses. Radiomics signatures of intratumor, peritumor, and habitat were developed using the features extracted from preoperative CT images. A nomogram was developed by integrating the optimal model and clinical predictors. The model performance and benefit were assessed using the area under the receiver operating characteristic curve (AUC), net reclassification index (NRI), and integrated discrimination improvement (IDI).

Results

A total of 382 eligible patients were included in the analysis. Two clinical variables (age and gender) were identified and used to construct the clinical model. The habitat-based radiomics model demonstrated superior discriminatory performance in differentiating PTC from NG, with AUCs of 0.948 (95% confidence interval [CI]: 0.923–0.973) and 0.941 (0.941, 95% CI: 0.896–0.985) in the training and validation sets, respectively. The combined radiomics nomogram achieved the highest predictive accuracy, with AUCs of 0.953 (95% CI: 0.930–0.976, training) and 0.950 (95% CI: 0.909–0.991, validation). Decision curve analysis (DCA) showed that the nomogram provided a higher net benefit than other radiomics models, supported by positive NRI and IDI values.

Conclusions

CT-based habitat radiomics had the potential to differentiate PTC from NG. The nomogram combined with Peri4mm and habitat signature had the best performance and good model gains for identifying PTC patients.