Endoscopic Image Analysis Using Deep Convolutional GAN And Traditional Data Augmentation

One big challenge encountered in the medical field is the availability of only limited annotated datasets for research. On the other hand, medical image annotation requires a lot of input from medical experts. It is noticed that machine learning and deep learning are producing better results in the area of image classification. However, these techniques require large training datasets, which is the major concern for medical image processing. Another issue is the unbalanced nature of the different classes of data, leading to the under-representation of some classes. Data augmentation has emerged as a good technique to deal with these challenges. In this work, we have applied traditional data augmentation and Generative Adversarial Network (GAN) on endoscopic esophagus images to increase the number of images for the training datasets, which have not yet been explored in literature. Eventually we have enhanced and applied the two deep learning models namely ResNet50 and VGG16 to extract and represent the relevant cancer features. The results show that the accuracy of the model increases with data augmentation and GAN. In fact, GAN has achieved the highest accuracy, that is, 94 % over non-augmented training set and traditional data augmentation for VGG16.