Advanced Image Processing for Breast Cancer Detection Using CNN-Based Transfer Learning on Mammograms

Breast cancer remains the most commonly diagnosed disease and the second leading
cause of death among females. Statistically speaking, roughly one out of every eight American
women was diagnosed with breast cancer last year. The precise identification of breast cancer also
largely relies upon careful analysis of medical images. Though several Deep Learning (DL)
algorithms have been employed to analyses such images, therefore, this study focuses on using a
Convolutional Neural Network (CNN) to differentiate between different types of mammograms.
The use of CNN in image recognition and visual processing has quickly drawn the attention of
scholars. Therefore, in this current research, an approach is presented to extract patches from
mammograms and utilize them to train the CNN, whereby the order of the section’s feeds into the
classification process. In addition, a transfer learning approach is utilized, in which a model created
in the initial phase is later utilized as an initial model. Besides using single and multi-CNN and
Artificial Neural Network (ANN) layers, two more approaches—Auto-Encoder and VGG16—are
used to evaluate and compare the effectiveness of the models on different datasets.