Data Augmentation-aided Convolutional Neural Network for Detection of Abnormalities in Digital Mammography
Olaide N. Oyelade , Ahmed Aminu Sambo , Pam Bulus Dung , and Amina Hassan Abubakar
Keywords: Breast cancer, convolutional neural network, deep learning, data augmentation, mammography,
Background: The use of data augmentation techniques to addressing the challenge of network overfitting and classification error is important in deep learning. Insufficient sample data for training have the tendency to bias the trained model so that it fails to generalize well. Several studies have proposed different augmentation techniques to solve this problem. But there are some peculiarities identified with the nature of datasets when applying augmentation methods. The subtle nature of some abnormalities in digital mammography often makes it difficult to transform such datasets into different form, while preserving the structure of the abnormality. Aim: To address this, this study aims to apply a combination of carefully selected data augmentation operations on digital mammography. Method: First, a convolutional neural network (CNN) is proposed suitable for feature extraction and abnormality detection from image samples. Secondly, transform operations are applied to the image data to generate more samples with different abnormalities to augment original datasets. Results: Using the image samples from the Mammographic Image Analysis Society (MIAS) with regions of interests (ROIs), Digital Database for Screening Mammography, Curated Breast Imaging Subset (DDSM+CBIS), INbreast, and whole images from MIAS, experimentation was carried out. The selected augmentation operations were applied to the datasets. These datasets contain samples presenting both bilateral craniocaudal (CC) and mediolateral oblique (MLO) views. Performance evaluation of the approach proposed in this study showed that classification accuracy of 90.62% was obtained with reduce loss values. The outcome of the study demonstrates the need to consider the structural and textural orientation of image samples when applying augmentation techniques to reduce overfitting.