Detection of Architectural Distortion with Deep Convolutional Neural Network and Data Augmentation of Limited Dataset

Early detection of breast cancer can increase treatment efficiency. One of the earliest signs of breast cancer is the Architectural Distortion (AD), which is a subtle contraction of the breast tissue, most of the time unnoticeable. A lot of techniques have been proposed over the years to aid the detection of AD in digital mammography but only a few using a deep learning approach. One of the most successful algorithms of deep neural architecture are the Convolutional Neural Networks (CNNs). However, to assure better CNN performance, the training step requires a large volume of data. This paper presents a deep CNN architecture designed for the automatic detection of AD in digital mammography images. For the training step, we considered the data augmentation approach, to overcome the limitation of clinical dataset. CNN performance was evaluated in terms of Receiver Operating Characteristic (ROC). The measured area under the ROC curve (AUC) was 0:87 for the proposed CNN in the task of AD detection in digital mammography.