Tongue contour tracking in dynamic ultrasound via higher-order MRFs and efficient fusion moves

We propose a graph-based method that required minimal user interaction. We employ a graph whose nodes represent vertices of segmentation contours in individual frames, and whose edges captured intra- (spatial) and inter-frame (temporal) relationships between contour vertices. We then formulate a graph-labelling problem, where each node is labelled with a displacement vector that maps it to a point on the reference contour. The optimal displacement labels are those minimizing a multi-label Markov random field energy with first and higher-order potentials that either captures image evidence or regularization constraints. [Display omitted] ► We propose a novel higher-order MRF model for 2D tongue contour tracking. ► We propose 2 schemes that wisely sample solution space for efficient optimization. ► We adapt temporal regularization according to contextual information. ► Evaluation on 63 sequences proved that our method outperforms previous methods. ► For reproducibility, we report all parameter settings and publicize our software at http://tonguetrack.cs.sfu.ca. Analyses of the human tongue motion as captured from 2D dynamic ultrasound data often requires segmentation of the mid-sagittal tongue contours. However, semi-automatic extraction of the tongue shape presents practical challenges. We approach this segmentation problem by proposing a novel higher-order Markov random field energy minimization framework. For efficient energy minimization, we propose two novel schemes to sample the solution space efficiently. To cope with the unpredictable tongue motion dynamics, we also propose to temporally adapt regularization based on contextual information. Unlike previous methods, we employ the latest optimization techniques to solve the tracking problem under one unified framework. Our method was validated on a set of 63 clinical data sequences, which allowed for comparative analyses with three other competing methods. Experimental results demonstrate that our method can segment sequences containing over 500 frames with mean accuracy of 3mm, approaching the accuracy of manual segmentations created by trained clinical observers.