A fast and automatic detection of epi- and endo-cardium using MR images

This paper proposes a fast and automatic algorithm to detect epi-and endo-cardium contours on short-axis cardiac magnetic resonance (CMR) images based on combined methods motivated from a double thresholding method. Briefly, it rapidly detects endo-cardium contour and left ventricular (LV) chamber region. Using the endo-cardium contour as initial seeds, the algorithm applies region growing of the LV in polar coordinates to achieve a reasonable epi-cardium contour. Preliminary results with the online cardiac MR database show that the proposed method can provide a mean absolute error of 54.325 ± 5.24 mm
²
for myocardium area computed from endo- and epi-cardium contour detection, accounting for 92% accuracy compared to the ground truth. The total time of detection is approximately 0.86 and 3.26 seconds per image frame for endo- and epi-cardium contours, respectively. These promising results reflect a potential use of the proposed algorithm for clinical applications in the future.