Understanding Clinical Mammographic Breast Density Assessment: a Deep Learning Perspective

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Oncology

Background:

• Mammographic breast density is an independent breast cancer risk marker.
• Current Breast Imaging and Reporting Data System (BI-RADS) density assessment relies on qualitative radiologist interpretation of MLO and CC views, exhibiting inter- and intra-reader variability.
• Understanding radiologist reading behaviors is crucial for improving consistency and accuracy in breast density assessment.

Purpose of the Study:

• To investigate how radiologists utilize MLO and CC mammogram views in BI-RADS density categorization using a deep learning approach.
• To assess the performance of a convolutional neural network (CNN) in classifying breast density categories from mammogram images.
• To identify which mammogram view (MLO or CC) is predominantly used for density assessment.

Main Methods:

• A CNN-based deep learning model was developed and trained on 15,415 real-world clinical mammogram images.
• The model was used to classify breast density categories based on MLO view images and CC view images separately.
• Performance was evaluated using the area under the receiver operating characteristic curve (AUC).

Main Results:

• The CNN model achieved significantly higher classification performance for breast density categories using MLO view images compared to CC view images (AUC comparison).
• This suggests that radiologists predominantly rely on the MLO view for determining BI-RADS breast density categories.
• The findings provide insights into radiologist reading characteristics.

Conclusions:

• The MLO view appears to be the primary view used by radiologists for breast density assessment.
• Deep learning models can effectively analyze mammograms and potentially interpret radiologist reading behaviors.
• This research can inform personalized clinical training to reduce reader variations in breast density assessment and improve breast cancer screening accuracy.