Context Enhanced Graphical Model for Object Localization in Medical Images

Abstract

AbstractObject localization is an important step common to many different medical applications. In this Chapter, we will review the challenges and recent approaches tackling this problem, and focus on the work by Song et.al. [20]. In [20], a new graphical model with additional contrast and interest-region potentials is designed, encoding the higher-order contextual information between regions, on the global and structural levels. A discriminative sparse-coding based interest-region detector is also integrated as one of the context prior in the graphical model. This object localization method is generally applicable to different medical imaging applications, in which the objects can be distinguished from the background mainly based on feature differences. Successful applications on two different medical imaging applications – lesion dissimilarity on thoracic PET-CT images and cell segmentation on microscopic images – are demonstrated in the experimental results.KeywordsImage RegionObject LocalizationSparse CodeInterest RegionCell SegmentationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.