Breast cancer detection and diagnosis using hybrid deep learning architecture

The detection of cancer regions in the breast is important to save the women patient's life. In this paper, mammogram image is classified into normal, benign, and malignant using the proposed hybrid Convolutional Neural Networks (CNN) architecture. The proposed system consists of a radon transform, data augmentation module, and hybrid CNN architecture. The radon transform transforms each spatial pixel in the source mammogram image into a time-frequency variation image. This image is data augmented to construct a new dataset from the existing dataset to improve the breast cancer detection rate. The data augmented images are classified into three different cases using the proposed hybrid CNN architecture. Further, a mathematical morphological-based segmentation algorithm is used to segment the cancer pixels. In this article, Mammographic Image Analysis Society (MIAS) and Digital Database for Screening Mammography (DDSM) datasets for mammogram image analysis are used to estimate the performance efficiency of the developed deep learning architecture. The developed CNN architecture provides 97.91% Se, 97.83% Sp, 98.44% Acc, and 98.57% JI on the mammogram images available in the DDSM dataset. The developed CNN architecture provides 98% Se, 98.66% Sp, 99.17% Acc, and 98.07% JI on the mammogram images available in the MIAS dataset. For both open access datasets, the experimental results are compared to recent similar works. From the extensive analysis of experimental results of the proposed method, the methodologies presented in this article clearly segment the boundary of the cancer region in an abnormal mammogram image.