BREAST CANCER CLASSIFICATION IN ULTRASOUND IMAGES USING TRANSFER LEARNING

Abstract

We explored three versions of a deep learning solution to computer-aided detection of ultrasound images of cancerous tumor tissues. Experimentally, our work proved that the pre-trained VGG16 model has the best outputs in the fine-tuned version. In short, our test accuracy ranges from 79% to 97%. We employed data augmentation to enlarge the amount of training data, and avoid overfitting. We have also employed the VGG16 pre-trained model, and added practical fine tuning to improve precision. This work offers a path into developing realistic and versatile deep learning frameworks for detecting breast cancer. The findings suggest that the fine-tuned model with pre-training medical data has increased the classification accuracy. These frameworks should complement and provide assistance for approaches of clinical diagnosis and treatment.