Automated left ventricle boundary delineation in left ventriculograms

Automated left ventricle (LV) boundary delineation from contrast ventriculograms (LVG) has been studied for decades. Unfortunately, no methods have ever been reported with sufficient accuracy evaluations. A new knowledge based multi-stage method to automatically delineate the LV boundary at end diastole (ED) and end systole (ES) is discussed here which has shown its robustness over a large LVG database. It makes extensive use of knowledge about LV shape and movement throughout its processing, including a regional pixel classification, a shape regression and a rejection classification. The method was trained and tested on a database of 375 studies whose ED and ES boundary have been manually traced. The cross-validated results are presented in the paper, showing that the accuracy was close to interobserver variability.