Deep convolutional neural network in medical image processing

Researchers have started constructing systems that could automatically analyze the medical images. In the initial phase (starting from 1970 to 1990), image processing was carried out using a sequential application of pixel analysis and mathematical model. By the end of 1990, supervised learning became popular in medical image processing, in which the trained images were used to design the models. But the conventional machine learning techniques require a crucial process of feature extraction to be carried out by the researchers. The next level of development is to construct the system, which can do feature learning automatically and can represent the images in an optimal way. Here lies the basis of many deep learning (DL) techniques in which the network comprises multiple layers that process the input image to output result by doing the feature extraction by itself. The most popular and efficient DL model for medical image processing is the convolutional neural network (CNN). There are various medical imaging modalities with the help of which critical images of the patient's body structures are captured and are fed to the CNN model for the analysis. In this chapter, authors have done a detailed discussion on different CNN architectures and their applications in the medical imaging domain. Moreover, a state-of-the-art comparison has been carried out between several existing works inside medical imaging based on CNN. Lastly, this chapter concludes with several critical remarks highlighting future challenges and their solutions.