Performance of a Fourier-based program for three-dimensional reconstruction of the mitral annulus on application to sparse, noisy data

Objectives: We investigated the accuracy of mitral annular reconstruction f rom noisy, sparse data typical of three-dimensional (3D) transthoracic echo cardiograms. Background: Our Fourier-based method for reconstructing the an nulus from dense, accurate 3D transesophageal echo (TEE) data has been vali dated in vitro with four harmonics in the x, y, and z coordinates (4,4,4). Methods: Thirteen mitral annuli were reconstructed from `complete' 3D TEE d ata using four harmonics (4,4,4) and used to measure area, eccentricity, he ight, perimeter, and interpeak and intervalley distances; these were the `t rue values'. To simulate transthoracic echo data, the TEE data sets were re duced evenly and unevenly (randomly). The complete and reduced data sets we re used to reconstruct the annuli using three sets of fitting parameters: ( 4,4,4), (1,1,3), and (1,1,4). The resulting size and shape measurements wer e compared with true values. Results: Regardless of the fitting parameters used, area, 2D perimeter, and 3D perimeter measurements were more accurate using reconstructions from evenly-reduced than randomly-reduced data sets ( p < 0.006), and depended significantly on both data density (p < 0.015 for all) and data distribution (p < 0.02 for all). Perimeter, height, and eccen tricity of the reconstructed annuli were more accurately measured using fou r harmonics (4,4,4). Conclusions: Mitral annuli can be reconstructed from s parse, noisy data using the (4,4,4) fit if at least 25 points are obtained from evenly distributed imaging planes. These results suggest that detailed analysis of mitral annular size and shape can be made accurately from 3D t ransthoracic echocardiograms.