An Optimized Transfer Learning and Deep Convolutional Neural Network Approach for Automated Breast Cancer Detection

One of the most common malignancies in the world is breast cancer, which primarily affects women, its early detection is important to diagnose the unwanted cell growth in the breast region. The survival rate increases to 90% when the growth is detected early. The objective of research is to identify different algorithms having a higher accuracy and is faster in detection of the malignant cell growth at the same time to assist doctors to detect the cancerous growth. The research intends to determine whether the existing Convolutional Neural Network architecture performs the detection accurately or not. In this approach of detection, we implemented various model architectures such as Deep-CNN, VGG16, VGG19, InceptionV3, MobileNetV2 on a vast dataset to learn from it. The various models were compared on the basis of their accuracy. VGG16 performed the best in comparison to every other model and came out to be the best of the implemented, afterwards various optimizers were applied on VGG16 to further increase the accuracy to find its best tuning parameter. The best accuracy achieved by tuning different optimizers of VGG16 was 97% when the optimizer was set to default, and the default optimizer was “Adam”. This accuracy over a real-life dataset is considered to be ideal for classification purposes.