Convolutional Neural Network Based Breast Cancer Risk Stratification Using a Mammographic Dataset

Area of Science:

• Artificial Intelligence
• Medical Imaging
• Oncology

Background:

• Mammography is a key tool for breast cancer screening.
• Accurate breast cancer risk assessment is crucial for early detection and prevention.
• Existing risk models may not fully leverage the detailed information within mammographic images.

Purpose of the Study:

• To develop and evaluate a novel pixel-wise breast cancer risk model using a convolutional neural network (CNN).
• To assess the predictive performance of the CNN model compared to breast density.
• To investigate the potential of AI in improving mammographic risk stratification.

Main Methods:

• A retrospective case-control study involving 1474 mammograms from average-risk women.
• Development of a fully convolutional CNN architecture for pixel-wise risk prediction.
• Training and validation of the CNN model using a dataset of women with and without breast cancer.

Main Results:

• The CNN model achieved 72% accuracy in predicting breast cancer risk.
• The CNN model demonstrated greater predictive potential (OR=4.42) than breast density (OR=1.67).
• Breast density was significantly higher in the case group, but the CNN model provided independent risk prediction.

Conclusions:

• A novel CNN-based pixel-wise mammographic risk model can effectively stratify breast cancer risk.
• This AI-driven approach offers a valuable tool for breast cancer risk assessment, independent of breast density.
• Further improvements are anticipated with larger datasets.