Adaptive Segmentation for Vessels Dynamic Characterization Using High Resolution MR Sequences

New challenge appeared recently in extracting mechanical hemodynamic properties from dynamic medical images. MR gated sequences may be sampled at high frequencies and enable precise observation of arteries pulse. We propose here an adaptive corrective thresholding (ACT) segmentation method to automatically extract arteries shape variations. It is robust even inside-vessel signal drops during RR period. ACT is able to approximate vessel contours using the last image in the time series as an elementary corrective structure. Furthermore, segmentation correction is adaptively performed considering vessel shape variations within the cardiac cycle. It is a fast and simple algorithm. For validation, the method was applied on a phantom and clinical data sets. Qualitative, quantitative and shape comparisons between manual and automatic results gave the segmentation performance evaluation. Results demonstrate that ACT overcomes movement-related artefacts with root mean square (RMS) error as low as 0.18 and 2 mm2 respectively on in vitro and in vivo images with maximal errors Emax of 0.81 and 5.44 mm2. The shape error (ERR) doesn't exceed 7%.