Robust Segmentation of Freehand Ultrasound Image Slices Using Gradient Vector Flow Fast Geometric Active Contours

We propose a new semi-automatic segmentation strategy on echocardiographic images, which combines a recently introduced gradient vector flow (GVF) fast geometric active contour (GAC) model and a modified level sets methods applied to echocardiographic data by Corsi et al.. We call it adaptive GVF GAC model. We note that echocardiographic images are characterized by high levels of speckle noise, weakly-defined boundaries and severe gaps. We show that the new method, adapted for single object segmentation, can provide significantly improved performance over a competing level set method, and that was in turn shown to perform better than the original gradient vector flow method. The new method modifies the advection term in the speed function adoptively by estimating how close the propagated curve is to the target boundaries. We show both synthetic and real, freehand ultrasound image and echocardiographic image examples to illustrate the robustness and accuracy of the new segmentation method