Abstract
This paper aims to provide a systematic review of the state of the art in the use of deep neural networks (DNNs) in medical imaging, an area that has been recently developed because of the emergence of artificial intelligence (AI) and machine learning (ML). Deep Neural Networks including Convolutional Neural Networks (CNNs) and Generative Adversarial Networks (GANs) have shown excellence in handling of a gigantic imaging data and assisting in diagnostics, treatment planning and patient care. This review also focuses on the breakthroughs of DNNs in different imaging tasks including classification, segmentation, registration, and detection, and has highlighted its potential to enhance diagnostic accuracy in different organs like the brain, lung and chest. Some of the key problems related to DNN deployment are also considered, including the problems that arise due to limitations of data, computational power, and model interpretability. That is why innovations such as transfer learning and synthetic data acquisition contributed to reducing these problems, thereby improving model performance with limited data. The paper concludes by discussing future works where the emphasis is made on the higher interpretability of the models and the combination of clinical records with images. In this paper, we attempt to offer a comprehensive review of the latest developments in DNNs in medical image analysis and delineate potential research directions to help researchers and practitioners interested in applying DNNs for medical imaging tasks.