Robust Automatic Knee MR Slice Positioning Through Redundant and Hierarchical Anatomy Detection

Abstract

Diagnostic magnetic resonance (MR) image quality is highly dependent on the position and orientation of the slice groups, due to the intrinsic high in-slice and low through-slice resolutions of MR imaging. Hence, the higher speed, accuracy, and reproducibility of automatic slice positioning, make it highly desirable over manual slice positioning. However, imaging artifacts, diseases, joint articulation, variations across ages and demographics as well as the extremely high performance requirements prevent state-of-the-art methods, such as volumetric registration, to be an off-the-shelf solution. In this paper, we address all these issues through an automatic slice positioning framework based on redundant and hierarchical learning. Our method has two hallmarks that are specifically designed to achieve high robustness and accuracy. 1) A redundant set of anatomy detectors are learned to provide local appearance cues. These detections are pruned and assembled according to a distributed anatomy model, which captures group-wise spatial configurations among anatomy primitives. This strategy brings about a high level of robustness and works even if a large portion of the target is distorted, missing, or occluded. 2) The detectors are learned and invoked in a hierarchical fashion, with each local detection scheduled and iterated according to its intrinsic invariance property. This iterative alignment process is shown to dramatically improve alignment accuracy. The proposed system is extensively validated on a large dataset including 744 clinical MR scans. Compared to state-of-the-art methods, our method exhibits superior performance in terms of robustness, accuracy, and reproducibility. The methodology is general and can be applied to other anatomies and other imaging modalities.

Detecting Anatomical Landmarks From Limited Medical Imaging Data Using Two-Stage Task-Oriented Deep Neural Networks

“…Although multi-atlas registration can improve the robustness and accuracy, it is very time intensive to perform multiple registrations for each testing subject. Different from the former two sets of methods, learning based methods utilize learning algorithms in machine learning domain for landmark detection and have demonstrated superiority in anatomical landmark detection for medical images [6], [19]. …”

“…Currently, there are a large number of classification based methods for localizing anatomical landmarks or organs [20], [19]. In these methods, voxels near a specific landmark are regarded as positive samples and the rest are used as negative ones.…”

Automatic Scan Planning for Magnetic Resonance Imaging of the Knee Joint

“…Finally, Zhan et al 19 suggest a completely different method based on a hierarchical redundant anatomy detection framework. This work has also been described in more detail and further evaluated in Zhan et al 18 However, all the three different methods presented so far have in common that they require a new and modified fully 3D scout imaging protocol for alignment, whereas the current manual alignment is based on 2D scout slices. The 3D scout image is not introduced in the clinics yet.…”

Online updating of context‐aware landmark detectors for prostate localization in daily treatment CT images

“…6,[26][27][28][29][30][31][32][33][34][35][36][37][38] For example, Criminisi et al 27 built landmark detectors upon randomized decision forests for detection and localization of anatomical structures in the CT volumes. Zhan et al 28 used a set of extended Haar-wavelet features and also the Adaboost learning method to train detectors for MR knee landmark detection. In the Criminisi's work, 38 they learned a regression model to detect anatomical structures in the CT images.…”

“…Note that, in this paper, each component of the displacement field is a displacement vector, which denotes the 3D displacement of each voxel in each image toward the corresponding target landmark on the same image. Since other landmark detection methods [26][27][28][29][30][31][32][33][34][35][36][37][38] cannot simultaneously fulfill these two types of spatial consistency for the estimated displacement fields, which are important for accurate landmark detection, their performance might be limited.…”