An integrated solution for semi-automatic segmentation of volumetric ultrasound data based on B-spline explicit active surfaces

Abstract

Real-time 3D echocardiography (RT3DE) is a noninvasive imaging modality used to provide diagnostic information about cardiac morphology and function. Unfortunately, manual analysis of these datasets remains cumbersome and time-consuming. Our team recently proposed a generic framework for automatic 3D segmentation of volumetric datasets (B-spline explicit active surfaces, BEAS). However, as fully automatic segmentation can (locally) fail, the option of manually correcting the segmentation result should be available for optimal clinical routine use. The aim of this study was to develop an interactive segmentation method by embedding the user input in the segmentation framework. In order to validate the proposed method, a database consisting of 10 3D echocardiographic images from open-chest sheep experiments was used. Two experts segmented the data, using a purely manual approach as well as the proposed interactive framework. Using the interactive approach, the RT3DE data could be analyzed in a more time efficient manner (analysis time, interactive: 44.7 ± 11.9s vs. manual: 181.5 ± 66.2s; p <; 0.001), while being an equally accurate alternative to manual contouring. Moreover, it improved the reproducibility of the extracted measurements (inter-observer variability, interactive: 3.1 ± 2.4ml vs. manual: 6.0 ± 2.7ml; p <; 0.05).