An Efficient Densely Connected Convolutional Networks based Detection and Classification of Breast Cancer using Mammogram Images

  • Remya R., Hema Rajini N.

Abstract

In the last decades, breast cancer becomes the second deadliest disease among women over the globe and mammograms can be utilized as an effective screening process to decrease the mortality rate. The identification of breast cancer in digital mammograms is tedious in the image classification process owing to the fact that the tumor occupies only a smaller region in the whole breast image. The latest research works found that the deep learning (DL) based computer aided diagnosis (CAD) model assists radiologists to diagnose breast cancer in an automated way. In this view, this paper introduces a new Densely Connected Convolutional Networks (DenseNet) based breast cancer diagnosis using mammograms. The proposed model incorporates several processes namely preprocessing, segmentation, feature extraction, and classification. For image segmentation, the Adaptive Kernel-Based Fuzzy C-Means Clustering (AKFCM) technique is employed and then the DenseNet model is utilized for feature extraction. Finally, a set of two classification models namely gradient boosting tree (GBT) and support vector machines (SVM) are included at the last layer of the DenseNet model called DN-GBT and DN-SVM models to identify the different class labels of breast cancer. The performance of the proposed model is determined using a benchmark dataset and the simulation outcomes are examined interms of diverse dimensions. The experimental outcome indicated that the proposed DN-GBT model has effectively diagnosed breast cancer with the maximum sensitivity of 85.25%, specificity of 96.43%, accuracy of 88.76%, and F-measure of 91.23%.

Published
2020-11-05
How to Cite
Remya R., Hema Rajini N. (2020). An Efficient Densely Connected Convolutional Networks based Detection and Classification of Breast Cancer using Mammogram Images. International Journal of Advanced Science and Technology, 29(04), 10663–10678. Retrieved from http://sersc.org/journals/index.php/IJAST/article/view/33578