To construct radiomics models using biplanar ultrasound images to differentiate non-mass breast carcinoma (NMBC) from mastitis and evaluate the diagnostic value of combining radiomics features with clinical variables.
Approach:
Study Design: Retrospective analysis of data from 139 patients (63 with NMBC and 76 with mastitis) using radiomics features extracted from transverse and longitudinal ultrasound images.
Model Development: Three logistic regression models were constructed for transverse, longitudinal, and fused imaging data, alongside clinical variable-based models and combined clinical-radiomics models.
Performance Assessment: Model performance was evaluated using area under the curve (AUC), accuracy, sensitivity, specificity, and decision curve analysis (DCA) values.
Key Findings:
The fusion model outperformed single-plane models in the training cohort (AUC = 94.2%, accuracy = 87.6%).
In the validation cohort, the performance of transverse, longitudinal, and fusion models was comparable (AUCs: 0.730, 0.823, 0.800; accuracies: 69.0%, 78.6%, 78.6%).
The combined clinical-radiomics model outperformed both radiomics and clinical variable-based models in the validation cohort (AUC: 0.861–0.884).
Interpretation:
Biplanar ultrasound imaging-based radiomics models show potential for distinguishing NMBC from mastitis, with combined clinical-radiomics models providing added diagnostic value.
Limitations:
The fusion model demonstrated limited generalizability in the validation cohort.
The study was retrospective and may be subject to selection bias.
Conclusion:
Integrating clinical variables with radiomics features may enhance diagnostic accuracy for differentiating NMBC from mastitis.
