ABSTRACT
Currently, innovations in deep learning and the availability of large annotated medical image datasets are leading to dramatic changes in automated learning of medical images. From this perspective, it is necessary to analyze how computer-aided diagnosis of medical images has evolved and how these advances in technology are going to impact on our society. Deep learning networks naturally integrate features and classifiers in a multilayered fashion, and the feature levels can be enriched by the number (depth) of stacked layers. Network depth is a crucial factor in analyzing the results. Since the successful implementation of ImageNet Classification, based on Convolutional Neural Networks (CNN), researchers have started actively participating in the rapidly growing field of deep learning. The major goal of this chapter is to give an overview of (1) the changes which have been made by the introduction of deep learning to healthcare, (2) the major deep learning models, and (3) the applications of deep learning models. Machine learning with feature input was dominant before the introduction of deep learning, and the essential difference between machine learning before and after the introduction of deep learning is the learning of image data directly without object segmentation, i.e., the source of the power of deep learning. There are two major models in the class of machine learning in medical imaging, massive training artificial neural networks (MTANN) and CNN, which have similarities as well as several differences. Appropriate usage of statistical methods in different learning models will help to achieve the easy identification of diseases. Advances in deep learning open a new era where it will be possible to automatically build features from the dataset. On the other hand, many important medical projects were launched during the past decade. The medical big data emerging from these projects can support the use of deep learning in medical diagnosis.
