Session SA1.1

An Automated Evaluation of Regional Left Ventricular Function on Cine MR Images

R El Berbari*, N Kachenoura, AB Redheuil, A Herment,
I Bloch, E Mousseaux, F Frouin

INSERM
Paris, France

The evaluation of the regional left ventricular (LV) function is essential to the diagnosis and the follow-up of coronary artery diseases. In clinical routine, it is mostly based on the visual interpretation of cine MR images. A novel approach combining a robust endocardial segmentation method [1] to a wall motion quantification process [2] is presented. It enables an automated estimation of functional parameters such as time of first contraction (Td), mean contraction time (Tm), and endocardial radial velocity (Vr).
The segmentation method is robust to the presence of papillary muscles and the flow heterogeneities inside the cavity. Short axis MR images were filtered using connected operators prior to the segmentation, which was performed with the Gradient Vector Flow - Snake algorithm. A first approximate contour of the whole cavity including the papillary muscles was obtained using a high value of the rigidity snake parameter. This contour was used as an initialization for a second segmentation step that was performed on the original image, with lower values of the rigidity parameter, in order to refine the contour and take into account small cavity boundaries details. This method was applied to end-diastolic images, and then combined to a quantification process [2] to derive parameters of regional LV function: Td, Tm and Vr. This whole process was applied to ten control subjects (age: 49±6) and ten patients (age: 57±13) with myocardial infarction (MI).
Results were focused on temporal parameters, since their estimation was very robust. The mean values ± standard deviations of Tm (respectively Td) were 356±53 ms (50±17 ms) in the control group, 396±60 ms (66±20 ms) in the MI group. Moreover a high correlation between heart rate and Tm was found in both groups (R>0.8). The Tm/R-R ratio was estimated to 0.40±0.3 in the control group and 0.44±0.4 in the MI group. It was the most discriminative parameter (p<0.002), and its highest regional values matched infarcted segments.
The only manual interaction is the definition of one point inside the LV cavity and the anterior intersection between the two ventricles. Thus, the segmentation method ensures an automated assessment of the regional LV function and thanks to its robustness; this approach may prove clinically useful.
[1] R. El Berbari et al. FIMH LNCS 4466, 2007, 453-62. [2] N. Kachenoura et al. J Magn Reson Imaging, 2007, 1127-32.

(Abstract Control Number: 265)