Automatic Labeling of Anatomical Structures in MR FastView Images Using a Statistical Atlas

Abstract

Abstract. We present a method for fast and automatic labeling of anatomical structures in MR FastView localizer images, which can be useful for automatic MR examination planning. FastView is a modern MR protocol, that provides larger planning fields of view than previously available with isotropic 3D resolution by scanning during continuous movement of the patient table. Hence, full 3D information is obtained within short acquisition time. Anatomical labeling is done by registering the images to a statistical atlas created from training image data beforehand. The statistical atlas consists of a statistical model of deformation and a statistical model of grey value appearance. It is generated by non-rigid registration and principal component analysis of the resulting deformation fields and registered images. Labeling of an unseen FastView image is done by non-rigid registration of the image to the statistical atlas and propagating the labels from the atlas to the image. In our implementation, the statistical models of deformation and appearance are both implemented on the GPU (graphics processing unit), which permits computing the atlas based labeling using GPU hardware acceleration. The running times of about 10 to 30 seconds are of the same magnitude as the image acquisition itself, which allows for practical usage in clinical MR routine.

Automatic Detection and Segmentation of Kidneys in 3D CT Images Using Random Forests

“…[6]) have often been used for this problem. However such approaches are subject to registration errors due to inter individual variability.…”

Regression Forests for Efficient Anatomy Detection and Localization in CT Studies

“…Our approach is fully probabilistic and, unlike previous techniques (e.g. [2,3]) maximizes the confidence of output predictions. The focus of this paper is both on accuracy of prediction and speed of execution, as we wish to achieve anatomy localization in seconds.…”

“…Although atlas-based methods have enjoyed much popularity [3,15,16] their conceptual simplicity is in contrast to the need for robust, cross-patient registration. Robustness is improved by multi-atlas techniques [17], at the price of slower algorithms involving multiple registrations.…”

Automatic recognition of anatomical regions in three-dimensional medical images

“…In contrast, the proposed approach applies such a post-processing method which ensures the correct and consistent recognition of the anatomical structures even if some of the regions are missing in the image. Most of the related publications [6], [7], [10], [11] assume the anatomy structure to be detected is always found in the image, and the presented methods were not tested in cases where this assumption is not true.…”

“…The most related approach [11] performs a non-rigid registration to align a statistical model to a full-body MRI scan, but its usability is limited to full-body MRI scans.…”