Deep Architecture for Breast Cancer Detection: Using Optimized VGG16 Model and Transfer Learning
Akash Singh Chauhan, Indrajeet Kumar, Kamred Udham Singh, Prabhdeep Singh · 2023
In this growing and advance era the world is suffering from many diseases and breast cancer comes at 3rd place when it comes to number of new cases estimated year year. Hence it has become significantly important for us to detect this deadly disease to prevent early demises of such patience. Since AI and ML have introduced many models which assist medical professionals in better decision making, we have experimented with various deep neural network models over BUSI dataset. In primary or first experiment we observed that model performed well when input data is normalized and split ratio is 90/10. That model produced 76.90% accuracy. In the next experiment various deep CNN with Transfer Learning [TL] have been tried and tested with the same learnings from previous experiment and VGG16 outperformed others and produced 78.21% accuracy. The learning from previous experiments have also been applied in third experiment while best performing model was further fined tuned with hyper-parameters like learning rate and Optimizers. Out of different learning rates and optimzers when the learning rate was set to. 0001 and optimizer was set to RMSprop, over VGG16 model. The model outperformed all the models constructed in this research work with 83.33% accuracy. We named it Optimized VGG16. Further the confusion matrix and ROC have also been drawn to validate and analyse our findings further.