Ultrasound Radiomics and Machine Learning Differentiate Granulomatous Lobular Mastitis from Breast Cancer

Overview

A combined machine learning model integrating ultrasound radiomics and clinical factors effectively distinguishes granulomatous lobular mastitis (GLM) from breast cancer preoperatively. The model demonstrated high diagnostic accuracy with AUCs of 0.935 in training and 0.833 in validation cohorts, supported by explainable feature contributions.

Background

Granulomatous lobular mastitis is a benign inflammatory breast condition often misdiagnosed as breast cancer due to overlapping ultrasound features. Accurate preoperative differentiation is critical as treatment strategies differ substantially between GLM and breast cancer. Conventional imaging modalities like mammography and MRI have limitations including cost, invasiveness, and diagnostic ambiguity. Machine learning applied to ultrasound radiomics offers a promising non-invasive approach to improve diagnostic precision.

Data Highlights

Key Findings

• The combined model integrating ultrasound radiomics and clinical predictors outperformed models based on radiomics or clinical data alone.
• Key radiomic features contributing to differentiation included lbp_3D_k_glszm_SmallAreaLowGrayLevelEmphasis and gradient_glcm_Imc2 among others.
• Decision curve analysis demonstrated favorable clinical utility of the combined model for preoperative diagnosis.
• Shapley additive explanations provided transparent interpretation of feature importance, enhancing model explainability.
• Ultrasound radiomics can non-invasively capture subtle imaging biomarkers distinguishing GLM from breast cancer.

Clinical Implications

This interpretable machine learning approach offers clinicians a non-invasive, accurate tool to differentiate GLM from breast cancer preoperatively, potentially reducing unnecessary invasive biopsies and guiding appropriate treatment strategies. Integration of radiomics with clinical factors enhances diagnostic confidence and supports personalized patient management.

Conclusion

The study demonstrates that a combined ultrasound radiomics and clinical factor model can effectively and transparently distinguish granulomatous lobular mastitis from breast cancer preoperatively, representing a valuable advancement in breast imaging diagnostics.

References

1. Quzhou People’s Hospital Study 2013-2023 -- Ultrasound Radiomics for GLM vs Breast Cancer