Dual view deep learning for enhanced breast cancer screening using mammography

Area of Science:

• Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Breast cancer is the most common cancer among Ethiopian women, often diagnosed late.
• Mammography screening is effective for early detection but requires expert radiologists, who are scarce in Ethiopia.

Purpose of the Study:

• To develop an AI model to aid radiologists in mass breast cancer screening.
• To improve early detection and patient prioritization for breast abnormalities.

Main Methods:

• An ensemble of EfficientNet-based classifiers was combined with YOLOv5 for suspicious mass detection.
• An abnormality detection model was integrated to enhance screening.
• The model aims to provide explanations for predictions and improve sensitivity.

Main Results:

• The classifier model achieved an F1-score of 0.87 and sensitivity of 0.82.
• Incorporating YOLOv5 suspicious mass detection increased sensitivity to 0.89.
• The F1-score slightly decreased to 0.79 with the addition of mass detection.

Conclusions:

• The developed AI model shows promise in assisting breast cancer screening in resource-limited settings like Ethiopia.
• Combining classification with object detection (YOLOv5) improves sensitivity for detecting abnormalities.
• Further refinement is needed to balance sensitivity and F1-score for optimal clinical utility.