Breast Cancer detection using Transfer Learning enabled CLAHE-ResIn model

Abstract

Backgrounds & Objective: Breast cancer is the most prevalent disease diagnosed in the women community. The mortality rate could be reduced significantly if this disease is diagnosed at an early stage. This paper focuses on development of a model for breast cancer prediction as an aiding model for women community.

Methods: Histopathological biopsy images have been used to detect tumor of breast. It is done by adapting pre-trained   framework of Deep Learning. It is also known as transfer learning.  Developed model receives an enhanced image as an input, find the region of interest, extract the features using convolutional neural network for better prediction of breast cancer. Images used with different magnification factors viz 40x, 100x, 200x, and 400x and Image enhancement is performed by Contrast Limited Adaptive Histogram Equalization (CLAHE) technique.

Results: The accuracy of this model for image with 400x magnification factor is 99.1%. This level of accuracy is better than some other existing models, which are using various hybrid approaches for breast cancer prediction.

Conclusion: Hybridization of Inception and Resnet architecture with image enhancement using CLAHE technique and Machine Learning (ML) classifiers provides better prediction than some other existing hybridization models. This difference in prediction results is due to missing of either CLAHE or ML classifier with deep learning in some existing hybridization models.

It is suggested that results may improve more by applying different other existing image enhancement techniques like Gaussian Filter etc. Different architecture of transfer learning may also be tested to improve the results.