Classification of invasive ductal carcinoma breast cancer using deep neural network

Abstract

Invasive Ductal Carcinoma (IDC) is a type of breast cancer that is common and it requires correct and early treatment. With technology advancement, automated computer tools have been developed for analyzing, diagnosing and predicting IDC. In this work, a study has been conducted to compare the performance of different methods of classifying images using Deep Neural Network. Two main approaches were used. Building a neural network model from scratch and using transfer learning. The methods had different input values. However, both approaches used Convolution Neural Network architecture. 3-way cross validation was used to split the datasets. The evaluation methods for analyzing the results were learning curves, confusion matrix, and classification report. The results showed that ResNet50 model had a better performance and had a lot of images that were correctly classified compared to the other methods with a total accuracy of 87%. This was in comparison of total accuracy of 84% and 85% achieved by a model built from scratch and Mobilenet respectively. The study went further to check the significance of color in IDC breast cancer image classification by comparing the performance of models with colored images and images on a gray scale. According to overall accuracy, precision and recall, CNN model from scratch and ResNet50 trained on colored images is performing better compared to CNN model from scratch and ResNet50 trained on grayscale images.