Optimal spatiotemporal Fourier filtration for noise reduction of echocardiographic endocardial borders

Quantitative left ventricular wall motion analysis by echocardiography is hindered by induced error in the localization of extracted endocardial borders. The authors postulate that this error is predominantly high frequency fluctuations about the true shape and motion. Endocardial borders from complete cardiac cycles were hand-digitized; Fourier transformed with respect to the two periodic variables, time and angle about the centroid; then reconstructed by inverse transformation after eliminating the high frequency noise components. Filtered contours appeared more physiologic than the raw data and yielded significantly more accurate ventricular diameter as assessed by simultaneously acquired M-modes. Spatiotemporal Fourier filtration should be applicable as well to representations of cardiac shape and motion derived from nuclear magnetic resonance, angiography, computed tomography, and nuclear techniques.