Automated Breast Density Assessment in MRI Using Deep Learning and Radiomics: Strategies for Reducing Inter-Observer...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Radiology
• Quantitative Imaging Biomarkers

Background:

• Accurate breast density evaluation is crucial for precise cancer risk estimation.
• High inter-observer variability in breast density assessment poses a significant challenge.

Purpose of the Study:

• To assess the feasibility of artificial intelligence (AI) in reducing inter-observer variability for breast density assessment.
• To evaluate AI-assisted interpretation for improving consistency in breast density classification.

Main Methods:

• A retrospective study involving 621 patients without breast prostheses or reconstructions.
• Development of deep learning and radiomics models for classifying breast density categories (BI-RADS A-D).
• AI-assisted interpretation through majority voting between models and radiologists on an independent test set.

Main Results:

• Radiologists showed substantial agreement for most breast density tasks, but moderate agreement for extremely dense categories.
• AI models demonstrated substantial agreement with the reference standard for most tasks.
• AI-assisted interpretation resulted in almost perfect inter-observer variability (mean kappa values of 0.86-0.94).

Conclusions:

• Deep learning and radiomics models show significant potential in reducing inter-observer variability in breast density assessment.
• AI-assisted interpretation can enhance the consistency and reliability of breast density evaluations.
• These AI tools may contribute to more standardized and accurate breast cancer risk stratification.