Automatic frame-to-frame contour detection in echocardiograms using motion estimation

Six techniques for motion estimation (matching) in echocardiographic image sequences were compared in combination with an automatic frame-to-frame contour detection method based on dynamic programming and resampling of the image around a contour model. The maching was performed on the resampled data arrays. For each array position, a match value was calculated and included in a cost matrix for the subsequent dynamic programming contour detection. The six matching methods compared were cross correlation (CRCOR), normalized cross correlation (NRCOR), correlation coefficients (CORCF), sum of absolute differences (SAD), sum of reciprocals of absolute differences (SRAD), and deterministic sign change (DSC). Results of matching were compared over ten series of 16 echo images. Best results were obtained with NRCOR, CORCF, and SAD. SAD was preferred for its speed.<
>