Deep Transfer Learning with Multi-Level Features Extraction Approach for Breast Cancer Classification

Breast cancer is a deadliest disease worldwide among the women. By some means early detection is very important to overcome it by prognosing on time. Deep networks having promising performance for different signal and image processing applications, breast cancer classification is one of the applications to diagnose efficiently. Deep transfer learning approaches are one of its kind and are using nowadays to overcome the computational time increase the performance as well. Sparse deep network such as VGG19 and ReasNet50 etc. are giving low performance but these are light weight and can be trained fastly. So, to increase the performance of the systems designed using these networks is an important issue, intermediate features are collected to overcome the vanishing gradient problem hence the performance can be improved. Here in this work histopathology images are used from which breast cancer is diagnosed. VGG19 deep network is considered as backbone network which is light weight, easy to implement and can be trained fast. Transfer learning is adapted to solve the problem of training the network from scratch. All the intermediate stage features are concatenated and finally these feature maps are feeded to fully connected deep network. Finally, classification is done with the help of SoftMax classifier. The proposed model is giving 96.43% accuracy which is better as compare to the existing state of the methods.