Computer Aided Detection of Mammographic Lesions Using Deep Learning

Inoue K, Kawasaki A, Koshimizu K, Yamanaka C, Doi T

doi:10.3804/jjabcs.26.204

Although the mammography screening system in Japan is managed by the Japan Central Organization on Quality Assurance of Breast Cancer Screening, system management varies regionally. The development of a novel screening method could improve system management, thereby increasing screening accuracy nationwide. The ”deep learning” approach to machine learning has recently been applied to many aspects of modern society. Here, we used a convolutional neural network（CNN）to evaluate the accuracy of mammography screening for the detection of breast cancer. A CNN with ten hidden layers, including convolutional layers and pooling layers, was programmed using the Python ver３.５ programming language and the Tensorflow library developed by Google. A total of １０４ mammography images that had been diagnosed as depicting breast cancer were automatically divided into ２０ image pieces, each with an image size of５１２x５１２ pixels. The CNN was then trained and tested using these ２０４６ images to evaluate the ability of the CNN to identify images containing breast cancer. The accuracy, sensitivity, specificity, positive prediction value, and negative prediction value were９４.９％,８８.５％, ９７.１％, ９１.１％ and ９６.１％, respectively, for the test data set. The deep learning approach to machine learning provided an excellent outcome in terms of the accuracy of detecting malignancies and is expected to contribute to the improvement and cost―effectiveness of mammography screening systems.

Computer Aided Detection of Mammographic Lesions Using Deep Learning

Bibliographic Information

Search this article

Abstract

Journal

References(3)*help

Keywords

Details 詳細情報について

Export

Report a problem