Identification of NPM and non-mass breast cancer based on radiological features and radiomics

Background Non-mass breast cancer, presenting with calcifications, asymmetric dense shadows, and architectural distortions, is challenging to distinguish from non-puerperal mastitis (NPM) due to radiological similarities on mammography. Purpose This study aims to develop a mammographic-based radiomics model to differentiate NPM from non-mass breast cancer, addressing the limitations of subjective BI-RADS assessments that risk misdiagnosis or delayed treatment. Methods Mammographic images from 104 patients (44 NPM, 60 non-mass breast cancer), collected from January 2018 to June 2023, were retrospectively analyzed. Two senior breast radiologists independently reviewed images, with disagreements resolved by a more senior radiologist. Regions of interest (ROIs) were manually delineated using 3DSlicer, and 576 radiomic features (shape, first-order, texture) were extracted using PyRadiomics. The Least Absolute Shrinkage and Selection Operator (LASSO) algorithm with 10-fold nested cross-validation selected 6 predictive features, and a support vector machine (SVM) model with a Radial Basis Function kernel was constructed. Performance was evaluated using nested cross-validation, calculating the area under the curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Results Calcification type and asymmetric dense shadows differed significantly between NPM and non-mass breast cancer (P < 0.05). The radiomics model achieved an AUC of 0.844 (95% CI: 0.787–0.904), accuracy of 0.769 (95% CI: 0.735–0.803), sensitivity of 0.883 (95% CI: 0.792–0.974), specificity of 0.678 (95% CI: 0.576–0.779), PPV of 0.784 (95% CI: 0.749–0.819), and NPV of 0.778 (95% CI: 0.662–0.896), compared with radiologists’ BI-RADS assessment (AUC: 0.860, 95% CI: 0.790–0.930; accuracy: 0.856, 95% CI: 0.787–0.923; sensitivity: 0.833, 95% CI: 0.736–0.926; specificity: 0.886, 95% CI: 0.791–0.979; PPV: 0.909, 95% CI: 0.832–0.984; NPV: 0.796, 95% CI: 0.679–0.907). Conclusions Radiomics using PyRadiomics-extracted features, LASSO, and SVM provides a robust quantitative tool to differentiate NPM from non-mass breast cancer, enhancing diagnostic precision and clinical decision-making.