[1P4-J-10] Vision, speech: organisms and medicine

[1P4-J-10-02] A novel quantification method of dense breast from mammography

〇Kenichi Inoue¹ (1. Shonan Memorial Hospital, Breast Cancer Center)

Keywords:breast screening, mammography, dense breast

[background] Evaluating dense breast from mammography is a challenging problem in term of objectiveness in breast screening. We developed a novel objective method evaluating dense breast by calculating the density within breast tissue area. [material and method] Mammography images classified as category 1 taken in our institute were collected. For each mammography image, masking image was created to indicate the are of the breast tissue. A total of 197 pair images were trained and tested with U-Net algorithm. Also, a “relative density” was calculated based on a “fat density” within a mammography. By summing the relative density within the breast tissue area, the “breast density” was calculated and evaluated whether each image was classified as dense breast. [result] DICE coefficient reached up to 87.0%. Defining a dense breast as the breast density being greater than 30%, 77% of the images were consistent with that evaluated by human. [conclusion] By using semantic segmentation, we developed a novel method calculating breast density and evaluating dense breast. Using this method, evaluation of dense breast at the breast screening becomes objective and stable.