Deep Convolutional Neural Networks: Foundations and Applications in Medical Imaging

Convolutional neural networks (CNNs) cause impressive improvements in dealing with text, speech, image, video, and other applications. This chapter covers the basic knowledge of CNNs. The developments of CNNs are discussed from different perspectives; specifically, the CNN design, activation function, loss function, regularization, optimization, normalization, and network depth, which reviews the benefits of each phase of CNN. The pre-trained CNNs and deep learning difficulties are presented. It also presents several used convolutional neural networks like AlexNet, VGG-19, GoogleNet, and ResNet. Furthermore, the chapter gives attention to different CNN types, which are region-based CNN, fully convolutional networks, and hybrid learning networks. The chapter summarizes the deep learning applications in medical imaging processing in terms of feature extraction, tumor detection, and tumor segmentation. Finally, a deep convolutional neural network (DCNN) is designed for brain tumor detection using MRI images. The proposed DCNN architecture is evaluated on the RIDER dataset achieving accurate detection accuracy within a time of 0.24 seconds per MRI image.