Abstract
Deep learning-based image segmentation is by now firmly established as a robust tool in image segmentation. It has been widely used to separate homogeneous areas as the first and critical component of diagnosis and treatment pipeline. In this article, we present a critical appraisal of popular methods that have employed deep-learning techniques for medical image segmentation. Moreover, we summarize the most common challenges incurred and suggest possible solutions.
Highlights
Medical image segmentation, identifying the pixels of organs or lesions from background medical images such as CT or MRI images, is one of the most challenging tasks in medical image analysis that is to deliver critical information about the shapes and volumes of these organs
To get some more benefits over other Convolutional Neural Networks (CNNs)-based systems, authors perfectly decided to fully utilize both local and global contextual features [43], coming from deeper and higher layers of network respectively. It is addressed by enhancing the model to multi-scale contextual that led to the construction of a very deep fully convolutional residual network (FCRN) consisting of 50 layers, to segment skin lesions with a Dice coefficient of 0.897 compared to 0.794 for the VGG-16
We first summarized the most popular network structures applied for medical image segmentation and highlighted their advantages over the ancestors
Summary
Medical image segmentation, identifying the pixels of organs or lesions from background medical images such as CT or MRI images, is one of the most challenging tasks in medical image analysis that is to deliver critical information about the shapes and volumes of these organs. One of the first pure 3D models was introduced to segment the brain tumor of arbitrary size [76] Their idea was followed by Kamnitsas [41] who developed a multiscale, dual-path 3D CNN, in which there were two parallel pathways with the same size of the receptive field, and the second pathway received the patches from a subsampled representation of the image. 3D max pooling is introduced which filters the maximum response in a small cubic neighborhood to stabilize the learned features against the local translation in 3D space This helped to achieve a much faster convergence speed compared to pure 3D CNN thanks to the application of the convolution masks with the same size of the input volume. This system has achieved 0.823 Dice score for lesion segmentation in CT images and 0.85 in MRI images
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have