Suever, Jonathan*

Project: Heart Tracking Software

Traditionally, clinicians have used classical measures of cardiac function such as ejection fraction to assess cardiac health; however, recent studies have shown that more advanced measures of cardiac mechanics (such as strain and torsion) are more predictive of mortality. Recent advancements in magnetic resonance imaging (MRI) allow us to image the motion of the heart and measure these advanced cardiac mechanics; however, these techniques require complicated acquisition and post-processing which make them unrealistic for routine clinical use. Conversely, cine steady state free precession (SSFP) images are acquired during any standard cardiac MRI examination to provide the physician with information regarding cardiac anatomy. We believe that we can utilize a machine learning based approach to measure tissue displacements and advanced cardiac mechanics from these routine cine SSFP images. This advancement would allow physicians, at any facility, to use cardiac mechanics to improve patient care.

Computational Methods:

We will employ a collaborative tracking approach which utilizes 1) an appearance model that is trained to track material points throughout the cardiac cycle 2) a shape model which enforces a realistic geometry and 3) a motion model which ensures temporal continuity. For the appearance model, we will train a set of appearance classifiers using an AdaBoost cascade. To generate the shape and motion models, we will use sparse manifold clustering. By combining these three components, we hope to accurately obtain displacements which can then be used to derive cardiac strain and torsion from routine cine SSFP images of the heart.

We hope to use the high performance computing resources available at the University of Kentucky to aid in the execution and parallelization of the appearance model training. We are currently prototyping our machine learning algorithms in Matlab with the long-term goal of transitioning to C++ and GPU-based methods.

Software:

Matlab
C++
CUDA

People:

Brandon K. Fornwalt, MD, PhD (Co-PI, UK CoM)
Lin Yang, PhD (Co-PI, UK CPH)
Jonathan D. Suever, PhD (Post-doc, CoM)
Fuyong Xing (UK CS PhD student)

Funding:

Currently unfunded

Publications:

None