Breast Cancer Detection Using Transfer Learning with DCGAN for dataset imbalance
Meyya Meyyappan, Aniket Verma, Ginikunta Sai Karthik Goud, Naga Malleswari TYJ, S. Ushasukhanya · 2024
Mammography is one of the primary procedures by which breast cancer detection is done, in order to provide early treatment and reduce mortality rate. With time Machine learning and Deep learning Frameworks were widely used to detect breast cancer. But there are certain limitations to it due the availability of the number of images available to train the model. This imbalance in the dataset leads to a major issue in implementing the model in real time. To solve this problem various data augmentations techniques have been used and also the use of Generative AI. By using Deep Convolutional Generative Adversarial Networks (DCGAN) new synthetic images can be created that looks realistic to new observers. The images generated using DCGAN are merged with the existing dataset to normalize the imbalance in the dataset. The balanced dataset is now used to train the transfer learning models like DENSENET201 and VGG16. After training both the models provided an improvement in accuracy from 63.5% to 64.16% when DENSENET201 model was used and 59.1% to 61% accuracy when VGG16 model was used. Dropout and Dense layers were added in then models to improve the model’s performance. The paper mainly focuses on preventing the limited availability of the dataset and the detection of tumor in breast.