Objective: This study aimed to explore the feasibility of habitat radiomics based on Non-Contrast Computed Tomography (NCCT) for differentiating Pleomorphic Adenoma (PA) and Adenolymphoma (AL), and to compare it with both clinical and conventional radiomics models. Methods: A retrospective collection of clinical and imaging data was conducted on 203 patients who underwent pathology-proven procedures from October 2015 to August 2024 at two hospitals. Tumor Regions of Interest (ROIs) were delineated on NCCT images, and the K-means algorithm was used to jointly cluster the training and validation sets. Radiomics features were extracted, followed by feature selection using the Minimal-Redundancy- Maximal-Relevance (mRMR) and Least Absolute Shrinkage and Selection Operator (LASSO) methods. Univariate and multivariate logistic regression analyses were conducted to identify clinical independent risk factors. The clinical, radiomics, and habitat models were constructed after selection of the clinical and radiomics features. The optimal radiomics model was combined with independent clinical risk factors to develop a nomogram and a combined diagnostic model. The performance of each model was evaluated using the Area Under the Receiver Operating Characteristic (ROC) Curve (AUC), and the DeLong test was used to compare model performance. Calibration curves and Decision Curve Analysis (DCA) were utilized to evaluate model calibration and clinical net benefit, respectively. Results: Four distinct habitat areas were identified through clustering analysis. The habitatₐll model achieved superior predictive performance, with AUCs of 0. 903 in the training set and 0. 846 in the validation set. This model outperformed the clinical model (training set AUC: 0. 837; validation set AUC: 0. 823), the conventional intra-tumor radiomics model (training set AUC: 0. 845; validation set AUC: 0. 840), and each of the four individual habitat models (training set AUCs: Habitat1 = 0. 839, Habitat2 = 0. 847, Habitat3 = 0. 822, Habitat4 = 0. 859; validation set AUCs: Habitat1 = 0. 823, Habitat2 = 0. 840, Habitat3 = 0. 827, Habitat4 = 0. 842). Furthermore, the nomogram integrating clinical independent risk factors (age and smoking history) with the habitatₐll model showed improved predictive performance (AUCs for the training and validation sets were 0. 953 and 0. 883, respectively) and demonstrated significant clinical net benefit. Conclusion: Habitat radiomics analysis based on NCCT enables accurate differentiation between PA and AL, providing novel insights for clinical diagnosis and treatment.
Liu et al. (2026) studied this question.