AI-Based Cancer Detection Model for Contrast-Enhanced Mammography

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Deep neural networks (DNNs) represent a breakthrough in computer-aided diagnostics (CAD) for breast imaging.
• Contrast-enhanced mammography (CEM) offers anatomical and functional insights but has limited deep learning (DL) research due to data scarcity.

Purpose of the Study:

• Develop and evaluate a CEM-CAD system for enhanced lesion detection and breast classification.
• Address the limited data availability for DL-based CEM analysis.

Main Methods:

• Optimized a YOLO-based deep learning model trained on a large CEM dataset (1673 patients, 7443 images).
• Evaluated lesion detection using FROC and breast classification using ROC metrics.
• Assessed performance with different image inputs and background parenchymal enhancement (BPE) levels.

Main Results:

• Achieved an AUROC of 0.964 for breast classification.
• Detecting 90% of cancers with a 0.128 false positive rate per image.
• Demonstrated superior performance using both low-energy and recombined images, with BPE significantly impacting results.

Conclusions:

• The developed CEM CAD system demonstrates high performance in lesion detection and breast classification.
• Its capabilities are comparable to those of experienced radiologists.