A semi-automatic endocardial border detection method for the left ventricle in 4D ultrasound data sets

Abstract We propose a semi-automatic endocardial border detection method for 3D+T cardiac ultrasound data based on pattern matching and dynamic programming, operating on 2D slices of the 3D+T data, for the estimation of LV volume, with minimal user interaction. It shows good correlations with MRI end-diastole (ED) and end-systole (ES) volumes ( r =0.940) and low interobserver variability ( y =1.005 x −16.7, r =0.943) over full-cycle volume estimations. It shows a high consistency in tracking the user-defined initial borders over space and time.