Contrast-Enhanced Mammography and Deep Learning-Derived Malignancy Scoring in Breast Cancer Molecular Subtype...

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Contrast-enhanced mammography (CEM) offers morphological and functional insights into breast cancer biology, comparable to Magnetic Resonance Imaging (MRI).
• Understanding CEM's ability to differentiate breast cancer subtypes is crucial for personalized treatment strategies.

Purpose of the Study:

• To evaluate if CEM imaging features and AI-derived malignancy scores correlate with molecular subtypes of breast cancer.
• To assess the diagnostic performance of AI in characterizing malignant lesions detected by CEM.

Main Methods:

• Retrospective analysis of 399 women with BI-RADS category 0 screening mammograms who underwent CEM.
• Analysis of 76 malignant lesions, classified by molecular subtypes (luminal, HER2-positive, triple-negative).
• Evaluation of imaging features (mass shape, enhancement) and deep learning-based AI malignancy scores (iCAD ProFound AI®).

Main Results:

• Luminal subtypes (69%) predominated; HER2-positive/triple-negative subtypes comprised 31%.
• CEM features like mass shape and enhancement patterns showed descriptive differences across subtypes.
• AI malignancy scores demonstrated good diagnostic performance (AUC=0.744) and were higher for malignant versus benign lesions.
• AI scores varied across subtypes, with higher median scores for luminal tumors, though not statistically significant.

Conclusions:

• CEM imaging characteristics align with known MRI-based phenotypes of breast cancer molecular subtypes.
• AI-enhanced CEM shows potential for improved lesion characterization and risk stratification in breast cancer diagnostics.
• Further evolution of AI models may enhance the clinical utility of CEM for personalized breast cancer management.