High resolution segmentation of MR images of mouse heart chambers based on a partial-pixel effect and EM algorithm

In this paper, a high resolution segmentation method for MR images of mouse heart chambers is developed, in which the EM algorithm is adopted and extended to the spatio-temporal cardiac cycle, and a partial pixel effect (PPE) is accommodated so that the heart chambers boundary is detected to sub-pixel accuracy. The method has been tested on high and low spatial resolution images. Quantitative measurements of the left ventricle (LV) area-time curve over cardiac cycle are reported, which show that the accuracy of the LV area can be improved based on the new method, on low spatial resolution images. The average relative error of the area estimate of the new method applied to low resolution images relative to measurements on high resolution images is less than 8%.