Session P84.5

Canine Left Ventricular Purkinje Fiber Network Construction Using Manifold Learning

J Li, KQ Wang*, WM Zuo, HG Zhang, YF Yuan

Harbin Institute of Technology
Harbin City , China

Purkinje fiber network, one of the most important components of the ventricular conduction system, is crucial in modeling ventricular tachycardia and fibrillation. Construction of anatomical detail Purkinje fiber network, however, is a very challenging task. Traditional medical imaging methods, such as magnetic resonance imaging (MRI) or computed tomography (CT), can be used to obtain the 3D structure information of the left ventricle, but fail to reveal the PFN information. In this paper, we present a method for restoring the 3D PNF in the left ventricle by manifold learning. Motivated by the fact that canine Purkinje fiber is generally on the endocardial surface of the heart, we have collected a set of 2D cut canine left ventricle images, extract the 2D location of the anatomical detail PFN, and then map the 2D PFN to the 3D canine left ventricular model released by Cornell University(http://thevirtualheart.org/). First, the PFN image is detected and extracted from the cut canine left ventricle image. To guarantee the extraction precision, we adopt a semi-automated method, where an automated image segmentation method is first used for PFN detection, and then manually post-processed to achieve the final satisfactory PFN image. To ensure that the local structure of the PFN is preserved, we first use a manifold learning approach for dimensionality reduction, resulting in that the original 3D data on the endocardial surface of the left ventricle is mapped to a 2D plane. We scale, rotate and translate the measurement of the plane surface to make it suitable to the PFN image. The discrete PFN image will be embedded on the 3D curved surface using the main bundle branches and boundaries as markers. When we successfully obtain a PFN-embedded plane surface image, we will use the inverse mapping to represent the 3D PFN which is within the 3D left ventricle muscle. Finally, anatomical detail 3D Purkinje fiber network is obtained with a satisfactory visible quality.

(Abstract Control Number: 161)