|
KMID : 1147120160220010001
|
|
Journal of the Korean Society of Imaging Informatics in Medicine
2016 Volume.22 No. 1 p.1 ~ p.8
|
|
|
Mammographic Breast Density Estimation using Deep Learning
|
|
Ahn Chul-Kyun
Kim Jong-Hyo
|
|
|
Abstract
|
|
|
Background : Deep learning has recently emerged as a powerful machine learning technique applicable to various pattern recognition problems including image recognition, speech recognition, data analysis, etc and has a good potential for use in medical image processing. We present our pilot study of applying convolution neural network (CNN) in automated mammographic breast density estimation.
Materials and Methods: 397 craniocaudal (CC) view full field digital mammograms were used in this study. Of the 397 mammograms, 297 mammograms were randomly selected as a training set and the rest 100 mammograms were used for a test set. Three experts determined boundaries between glandular and fat areas, from which square-shaped patches were extracted from each of glandular and fat area. Those patches were fed into the CNN for training.
Results: The trained-CNN produced patch-based segmentation of mammograms, which in turn gave percent mammographic density for the 100 test mammograms. The correlation between CNN-derived breast density and manually measured density by three experts was 0.77.
Conclusion: We were able to train CNN for estimating breast density from mammograms. As the deep learning technique has an ability to improve the performance by providing appropriate training with large amount of data, we expect a growing performance of deep learning technique with extended experience and accumulation of open medical imaging database.
|
|
|
KEYWORD
|
|
|
Deep learning, Convolutional neural network (CNN), Mammography breast density, Quantitative measure
|
|
|
FullTexts / Linksout information
|
|
|
|
|
|
Listed journal information
|
|
|
|