Cardiac function quantification from multislice computerized tomography images

The aim of this paper is to propose a technique to estimate several descriptors associated with the left ventricular function. The 3-D automatic segmentation of left ventricle (LV), in multi-slice CT cardiac images, was generated. The estimated descriptors were: end-diastolic volume, end-systolic volume, stroke volume, ejection fraction and cardiac output. Each considered image was processed using the following stages: preprocessing, LV segmentation and descriptors estimation. During the preprocessing stage a similarity enhancement based on a filtering technique, was applied and a learning paradigm for estimating two planes that isolate the LV from surrounding structures was used. The automatic LV segmentation was generated using a level set algorithm. The aforementioned descriptors were estimated from the LV segmentation. The percentage relative error was used to compare the results obtained with respect to those generated manually by a cardiologist. The descriptors have errors less than 5%, however, it is suggested to perform a more complete validation.