Detection of Breast Cancer from Mammograms using a Hybrid Approach of Deep Learning and Linear Classification

Posing a threat to a large portion of the population on a worldwide scale, breast cancer is now a deleterious pandemic. Chances of survival greatly improve if detected early, giving us impetus to contribute to various methods of detection from a technical perspective. A tool used to detect and diagnose breast-related diseases is a digital mammogram. The main purpose of this paper is to detect the probability of early stage breast cancer detection using mammography images. These mammograms were pre-processed using CLAHE technique. In order to detect masses from the aforementioned mammograms, a hybrid approach has been employed: combining a neural network with a linear classifier. Deep Learning model VGG16, based on convolutional neural networks which are efficient for image processing, was used for feature extraction, which were then fed into linear classifiers. Linear classifiers are also efficient, especially for rich data in terms of training and testing. Using this approach, the results of the mammogram were classified as normal or abnormal, with normal indicating that no tumour was present, abnormal indicating that a tumour (benign or malignant), calcifications, circumscribed, spiculated or other masses, distortions, or asymmetry may be present. This method was fairly successful in classifying mammograms as normal or abnormal correctly.