Towards automated analysis in 3D cardiac MR imaging

Abstract

The main purpose of the work described in this paper is to make a first step towards the automatization of the quantification of the ventricular volume in systole and diastole using MR Images.

To achieve this result, we pursue three partial objectives:

• 1. Obtain objective image segmentation. Manual organ delineations vary from physician to physician. An automatic segmentation, taking into account the typical nature of cardiac MR Images, should produce objective and reproducible results and take less time than manual segmentation.

• 2. Obtain reliable segment labeling. A computer system, which takes into account descriptions of the organs (scene knowledge), has to be developed to assist the physician in labeling the segments produced by the automatic segmentation. Interactive tools should be provided to show the results of segmentation and labeling to the physician and ask him for confirmation or corrections.

• 3. Obtain accurate volume measurement. Volume measurements will allow the evaluation and partial validation of the results obtained by the previous parts of the system.

The paper describes a prototype of a complete system for cardiac volume estimation. Detailed descriptions of the individual segmentation and labeling modules have been published previously. In this paper the emphasis lies on the interaction between these modules, their performances in the system, their 3D generalisation, and their evaluation based on cardiac volume estimations.