Calculation of the ejection fraction from cardiac cine MR images requires segmenting multiple images of the left ventricle. This process, which is often performed manually, is time-consuming and observer-dependent. In this work, an unsupervised machine learning algorithm, combining hidden Markov random field and optical flow, has been proposed to perform semi-automatic tissue segmentation on T1/T2-weighted low-rank tensor images that have a built-in feature space due to low-rank factorization performed during image reconstruction. The segmentation results then allow automatic EF calculation. Demonstrated results have higher efficiency and similar accuracy compared with manual segmentation, and were stable with respect to different initializations.