Database guided detection of anatomical landmark points in 3D images of the heart

Automated landmark detection may prove invaluable in the analysis of real-time three-dimensional (3D) echocardiograms. By detecting 3D anatomical landmark points, the standard anatomical views can be extracted automatically in apically acquired 3D ultrasound images of the left ventricle, for better standardization of visualization and objective diagnosis. Furthermore, the landmarks can serve as an initialization for other analysis methods, such as segmentation. The described algorithm applies landmark detection in perpendicular planes of the 3D dataset. The landmark detection exploits a large database of expert annotated images, using an extensive set of Haar features for fast classification. The detection is performed using two cascades of Adaboost classifiers in a coarse to fine scheme. The method is evaluated by measuring the distance of detected and manually indicated landmark points in 25 patients. The method can detect landmarks accurately in the four-chamber (apex: 7.9±7.1mm, septal mitral valve point: 5.6±2.7mm; lateral mitral valve point: 4.0±2.6mm) and two-chamber view (apex: 7.1±6.7mm, anterior mitral valve point: 5.8±3.5mm, inferior mitral valve point: 4.5±3.1mm). The results compare well to those reported by others.