Breast density semantic segmentation and quantitative evaluation using deep learning.

What is it about?

Breast density screenings are an accepted means to determine a patient's predisposed risk of breast cancer development. After examining a patient's mammogram images, the radiologist will assign a patient with one of four BI-RADS (Breast Imaging Reporting and Data System) breast density classes. These screenings are typically subjective and results can vary between radiologists. We outline an approach that can be used to develop a DeepLabV3-based breast density semantic segmentation model. After development, this model can be used to calculate a density metric along with a linear scale, a probability scale. Alongside these metrics, a visual representation of the model's output can be provided to aid in the interpretation of the results. These quantitative values can then be used to help standardize and objectively evaluate a patient's breast density.

Featured Image

Photo by National Cancer Institute on Unsplash

Photo by National Cancer Institute on Unsplash

Why is it important?

The variability in breast density screening results among radiologists can impact a patient's determined breast density class and their recommended treatment plan. Our results show that the use of a breast density semantic segmentation model can be used to provide radiologists with objective and quantitative metrics outside of just a breast density classification. These quantitative metrics can aid in the reduction of inter-rater variability between different radiologists and the visualization of the model's output can help with the interpretation of the results.

Perspectives