Using data augmentation techniques on dermoscopic images to improve the accuracy of the convolutional neural network

Abstract

Computer vision and machine learning techniques capable of assisting specialists in different areas in their day-to-day tasks are the subject of several studies. For example, in the area of health, there is the diagnostic aid system that focuses on diagnosing certain diseases early. This early diagnosis is very important for improving patients' quality of life. One of the main fields using these techniques is the classification and detection of objects in images using convolutional neural networks. It is worth noting that when developing applications using deep learning models, large volumes of data are needed to train the networks. And one of the major problems is the difficulty of obtaining a data set large enough to adequately train convolutional neural networks. One way around this problem is to create synthetic data from the available images, i.e. to apply data augmentation techniques. In this work, data augmentation techniques will be applied to improve the classification accuracy of skin lesion images using a convolutional neural network. At the end of this work, it is hoped that the techniques applied can be used as inspiration for other diagnostic aid systems and also to improve existing applications in the medical field.