A combined deep CNN-lasso regression feature fusion and classification of MLO and CC view mammogram image

Breast cancer is the most frequent disease among women, and it is a serious threat to their lives and well-being. Due to high population expansion, automatic mammography detection has recently become a critical concern in the medical industry. The emergence of computer-assisted systems has aided radiologists in making accurate breast cancer diagnoses. An automated detection and classification system should be implemented to prevent breast cancer from spreading. Breast densities vary widely among women, which causes missed cancers. In the case of breast density, the deep CNN algorithms can significantly reduce radiologist workload and improve risk assessment. The goal of this paper is to offer a deep learning strategy for identifying MLO and CC views of breast cancer as malignant, benign, or normal using an integration of deep convolutional neural networks (CNN) and feature fusion of LASSO (Least Absolute Shrinkage and Selection Operator) regression. The proposed method comprises pre-processing, data augmentation, feature extraction, feature fusion, and classification. The generated features were fed into LASSO regression for the best prediction in this system, which utilized CNN for feature extraction. The fused features were then transferred to CNN's fully connected layer for mammography classification. In our experiment, the publically available dataset CBIS-DDSM (Curated Breast Imaging Subset of DDSM) was employed. The proposed work gained an accuracy of 99.2%, specificity of 98.7%, AUC of 99.8%, sensitivity of 99.4%, and FI-score of 98.7%, which is higher than multi view CNN without a feature fusion based system.