The computer-assisted analysis for better interpreting images have been longstanding issues in the medical imaging field. On the image-understanding front, recent advances in machine learning, especially, in the way of deep learning, have made a big leap to help identify, classify, and quantify patterns in medical images. Specifically, exploiting hierarchical feature representations learned solely from data, instead of handcrafted features mostly designed based on domain-specific knowledge, lies at the core of the advances. In that way, deep learning is rapidly proving to be the state-of-the-art foundation, achieving enhanced performances in various medical applications. In this article, we introduce the fundamentals of deep learning methods; review their successes to image registration, anatomical/cell structures detection, tissue segmentation, computer-aided disease diagnosis or prognosis, and so on. We conclude by raising research issues and suggesting future directions for further improvements.

BackgroundStudies for infants are usually hindered by the insufficient image contrast, especially for neonates. Prior knowledge, in the form of atlas, can provide additional guidance for the data processing such as spatial normalization, label propagation, and tissue segmentation. Although it is highly desired, there is currently no such infant atlas which caters for all these applications. The reason may be largely due to the dramatic early brain development, image processing difficulties, and the need of a large sample size.MethodologyTo this end, after several years of subject recruitment and data acquisition, we have collected a unique longitudinal dataset, involving 95 normal infants (56 males and 39 females) with MRI scanned at 3 ages, i.e., neonate, 1-year-old, and 2-year-old. State-of-the-art MR image segmentation and registration techniques were employed, to construct which include the templates (grayscale average images), tissue probability maps (TPMs), and brain parcellation maps (i.e., meaningful anatomical regions of interest) for each age group. In addition, the longitudinal correspondences between age-specific atlases were also obtained. Experiments of typical infant applications validated that the proposed atlas outperformed other atlases and is hence very useful for infant-related studies.ConclusionsWe expect that the proposed infant 0–1–2 brain atlases would be significantly conducive to structural and functional studies of the infant brains. These atlases are publicly available in our website, http://bric.unc.edu/ideagroup/free-softwares/.

No abstract

No abstract

Modeling relation between actors is important for recognizing group activity in a multi-person scene. This paper aims at learning discriminative relation between actors efficiently using deep models. To this end, we propose to build a flexible and efficient Actor Relation Graph (ARG) to simultaneously capture the appearance and position relation between actors. Thanks to the Graph Convolutional Network, the connections in ARG could be automatically learned from group activity videos in an end-toend manner, and the inference on ARG could be efficiently performed with standard matrix operations. Furthermore, in practice, we come up with two variants to sparsify ARG for more effective modeling in videos: spatially localized ARG and temporal randomized ARG. We perform extensive experiments on two standard group activity recognition datasets: the Volleyball dataset and the Collective Activity dataset, where state-of-the-art performance is achieved on both datasets. We also visualize the learned actor graphs and relation features, which demonstrate that the proposed ARG is able to capture the discriminative relation information for group activity recognition. 1