Radiomics-based differentiation of benign and malignant breast masses on contrast-enhanced mammography: a...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Breast cancer diagnosis relies on accurate differentiation of benign and malignant masses.
• Contrast-enhanced mammography (CEM) provides valuable information for breast lesion characterization.
• Radiomics offers a quantitative approach to extract imaging features for diagnostic purposes.

Purpose of the Study:

• To differentiate benign and malignant breast masses using radiomic features from low-energy and recombined CEM images.
• To evaluate the diagnostic performance of various machine learning classifiers for breast mass classification.
• To assess the utility of open-source tools in a radiomics workflow for CEM analysis.

Main Methods:

• Retrospective analysis of 145 patients who underwent CEM.
• Extraction of radiomic features from manually segmented regions of interest on low-energy and recombined CEM images.
• Utilized an open-source workflow (ITK-SNAP, PyRadiomics) and machine learning classifiers with 10-fold cross-validation and an independent test set.

Main Results:

• Ensemble learning achieved the highest diagnostic performance for both image types.
• Recombined CEM images with ensemble learning yielded an accuracy of 91.8% and AUC of 0.978.
• High sensitivity and specificity were observed for different classifiers, with ensemble learning and neural networks showing top performance.

Conclusions:

• Radiomics analysis of CEM images effectively differentiates benign and malignant breast masses, potentially improving diagnostic accuracy.
• A radiomics workflow using recombined CEM images and open-source tools can complement conventional interpretation and aid in non-invasive lesion characterization.
• This approach supports the development of decision-support tools for clinical breast imaging applications.