A Novel Approach to Classify Breast Cancer Tumors Using Deep Learning Approach and Resulting Most Accurate Magnification Factor

Abstract

In the recent research, breast cancer has come out to be the biggest reason behind death among females. Detection of breast cancer in its earlier stages is really the need of time. Detection of cancerous tumor is really a long and time taking process which is very costly and requires a lot of workforce as well. Therefore, providing a computer assisted approach for the simpler classification areas could be a solution. This could simplify the complex examine process of pathologists. Hence, the most appropriate magnification factor based CNN framework is proposed which leads to lesser expenses and lesser workforce. The proposed framework considers the different magnification factor images to obtain the classification accuracy. The proposed CNN-based framework is sectioned into three parts: preprocessing, feature extraction based on CNN, and CNN-based classification. The preprocessing phase consists of four processes, i.e., resha**, formatting, image labeling, and train–test splitting. These preprocessings are applied after training and testing procedure of CNN model. In experimental analysis, publicly available breast cancer dataset from histopathology is used. The dataset consists of seven thousand nine hundred and nine images obtained from 82 different patients. These images are distributed among four different magnification factors: 40X, 100X, 200X, and 400X. These magnification factors are utilized for training of CNN model and for obtaining the accuracy. In result analysis, four different experimentations are conducted to show the effectualness of the presented CNN-based framework. Also, the results exhibit that the CNN-based framework used in this work achieves the highest average accuracy of 97.63% among the other architectures and existing approaches.