Vector Motion Analysis for Objective, Quantitative Analysis of Echocardiograms: First Large-Scale Clinical Validation in 100 Patients
M. Sühling, C.H.P. Jansen, S.V. Aschkenasy, P. Buser, S. Marsch, M. Unser, P.R. Hunziker
Gemeinsame Jahrestagung der Schweizerischen Gesellschaft für Kardiologie, der Schweizerischen Gesellschaft für Angiologie, Basel BS, Swiss Confederation, June 2-4, 2004, vol. 7, supp. 6, pp. 26 S.
Recently, the novel technique of Multiscale Motion Mapping has been added to the echocardiographic toolbox. Based on computer vision technology, it allows detection of motion in conventional grayscale images. In contrast to tissue Doppler techniques, this technique delivers true 2D motion vectors, is not limited to motion parallel to the ultrasound beam and can be performed on conventional ultrasound machines. To assess the value of this technique in clinical practice, we applied this tool to 100 clinical routine echocardiograms to assess its feasibility and diagnostic usefulness.
Methods: Echocardiographic sequences were digitally stored in standard views. A total of 1200 segments was analysed by three independent, blinded observers in a conventional manner for interobserver agreement and diagnostic classification. In parallel, segments were quantitatively analyzed by two observers by multiscale motion mapping for quantitative and qualitative inter- and intraobserver analysis. An artificial neural network was trained with the data acquired with the quantitative algorithm using 50% of the data. Then the other 50% of the data were used as a test set for the neural network. A majority voting approach was chosen to determine normal versus abnormal motion.
Results: Automated, quantitative analysis was feasible was 95% of patients. In a measured range of -1cm to 7cm radial systolic velocities, there was no significant inter- or intraobserver bias; standard deviation of measurement for Bland-Altmann analysis was 0.56 (inter-) and 0.85 (inter-observer). Pairwise interobserver agreement in subjective analysis measured by Cohen's Kappa was 0.46, 0.46, and 0.68 for classification of overall wall motion as normal or abnormal; agreement between automated analysis by multiscale motion mapping and subjective analysis was in a similar range (kappa of 0.53). During the study, it was also noted that reading echos with multiscale motion made the observers aware of subtle abnormalities that were otherwise not noted, but clearly there in a second look.
Conclusion: Multiscale motion mapping, a novel tool for determination of complex motion in clinical echocardiograms, is feasible in the large majority of patients. It yields quantitative motion vectors, is reproducible, and stimulates the user to increased awareness for regional wall motion.
@INPROCEEDINGS(http://bigwww.epfl.ch/publications/suehling0403.html,
AUTHOR="S{\"{u}}hling, M. and Jansen, C.H.P. and Aschkenasy, S.V. and
Buser, P. and Marsch, S. and Unser, M. and Hunziker, P.R.",
TITLE="Vector Motion Analysis for Objective, Quantitative Analysis of
Echocardiograms: {F}irst Large-Scale Clinical Validation in $100$
Patients",
BOOKTITLE="Gemeinsame Jahrestagung der Schweizerischen Gesellschaft
f{\"{u}}r Kardiologie, der Schweizerischen Gesellschaft f{\"{u}}r
Angiologie",
YEAR="2004",
editor="",
volume="7, Supplementum 6",
series="",
pages="26 S",
address="Basel BS, Swiss Confederation",
month="June 2-4,",
organization="",
publisher="",
note="")