Comparing Heart Models and Locating Anatomical Features

John A. Bogovic

Supervised by Drs. R. Martin Arthur and Jason W. Trobaugh
Department of Electrical and Systems Engineering
Washington University in St. Louis
Spring 2005

Abstract

Electrocardiograms can be used to infer the electrical behavior of the heart, and thereby estimate a patient’s risk of having an arrhythmia.  Using individualized heart models can improve the accuracy with which we can estimate the condition of the heart and assign risk.  We considered the possibility of developing a template heart based on the Visible Human male (see The Visible Human Project), which we could scale, translate and rotate to match the patient’s heart as closely as possible.  Furthermore, we have developed methods of visualizing the 3d Visible Human image data from several orthogonal viewpoints.  We are able to step through the structure one slice at a time, locating anatomical features of interest and including them in our template.

 

Heart Comparison

Two heart models were compared by computing the distance from each point on one heart to the closest point on the other (implemented by Michael Attig).  Similar results were obtained by comparing two triangulated surfaces - the midpoints of each triangle were compared.  We could then determine the anatomical locations that contributed the most error.  Furthermore, we can minimize the error at certain locations by aligning the areas by translating the hearts (see graphs below). 

 

Graphs of the same heart, aligned differently, showing the difference between it and another model heart.

Error Legend
	Above, the hearts were aligned at their centers.	Above, the hearts were aligned at the apex. (Note the smaller error there)

 

Moving Forward

Currently, we deform the template heart based on eyeball approximation: yielding an average error of about 2cm.  This is a promising result considering the considerable difference between the heart models and the simple method we used to scale and translate the heart model.  The error can be improved using an optimization algorithm that rotates and scales in 3 dimensions independently (stretches) the heart model while minimizing a value of interest.  We may focus on minimizing the average error, or error around a certain anatomical feature.  Which we choose will depend on how much the deformed heart template improves our estimation of the heart’s electrical behavior.   

Locating Anatomy

Images of the x-y plane from the Visible Human were first compiled into an image array.  By plotting pixels from multiple x-y plane images with the same x or y values, we can view the color data as images in the x-z or y-z planes.  Images from any plane can be viewed in sequence, allowing one to "step through" the heart.  As shown below, the color data can be sliced by orthogonal planes in 3D space and plotted together.   Viewing the heart in 3D allows us to locate anatomy more clearly than viewing the images in one plane at a time.