Enhanced Intraplaque Hemorrhage Delineation Method in Slab-selection Phase-sensitive Inversion-recovery (SPI) Sequence with MRI

Introduction: Atherosclerosis is the number one killer in the world. Previous histopathological and prospective studies [1,2] have shown that intraplaque hemorrhage (IPH) into the carotid atherosclerotic plaque is a major factor causing plaque instability and progression. Thus, IPH is critical to evaluation of carotid atherosclerotic disease. Among techniques proposed for in vivo IPH imaging and evaluation , the Slab-selection Phasesensitive Inversion-recovery (SPI) MRI sequence is reported to show improved IPH evaluation compared to others due to better IPH contrast. Using this imaging technique, a mean-shift model based automated IPH detection algorithm was developed which shows high correlation with human evaluation on slices with IPH in in vivo MR images . However, there are some limitations: 1) when the mean-shift algorithm is applied to SPI slices without IPH, it may be over sensitive and mis-identify IPH regions; 2) for IPH regions with small size or low contrast, mean-shift usually excludes the boundary pixels and under-estimates IPH area measurements. This is caused by the radius of distribution cluster in mean-shift model, which cannot be adjusted rapidly enough during the searching process when encountering MRI images with extensive intensity ranges or that contains IPH with small size and low contrast. Purpose: 1) develop enhance mean-shift model (eMS) to overcome the existing limitations; 2) evaluate the performance of the eMS algorithm and compare with the previous mean-shift based IPH detect algorithm by using human analysis results as ground truth. Methods and Materials: Study protocol: in this study, five subjects with documented carotid atherosclerotic plaque were recruited after informed consent. These subjects were scanned in a 3T clinical scanner (Philips Achieva, R2.6.1, Best, and the Netherlands) with the SPI protocol. For each subject, 32 slices were obtained with the SPI sequence. Detailed imaging parameters were: IRTFE sequence, TR/TE 13.2/3.2 ms, TI 400ms, FOV 160x160x32, voxel size: 0.6x0.6x2mm, reconstructed voxel size: 0.3x0.3x1mm, TFE factor 40, Phase-sensitive reconstruction, imaging time: 2m57s. Enhanced Mean-shift Delineation Algorithm: With the SPI technique, IPH can usually be accurately identified by mean-shift model on the slices with IPH due to its higher signal than other regions in the vessel wall [6] because IPH pixels are classified as a very distinct cluster in signal space within mean-shift model. However, in slices without IPH, the intensity dynamic range decreases greatly. If the cluster radius during mean-shift searching process cannot change accordingly, vessel wall area is likely over-segmented which can mislead the follow up identification. To overcome this problem, a weighting parameter is introduced in the mean-shift model which dynamically adjusts radius size in each iteration by evaluating the region size and signal features, such as the contrast in each region. With this design, the mean-shift model can intelligently adjust its segmentation scale during the iteration process. The iterative searching process uses the difference between local mean and center of distribution as criteria, which