In the United States, cardiovascular diseases are among the leading causes of persistent health problems, especially among the elderly. Early detection of potential heart problems can dramatically improve a patient's prognosis. One diagnostic method of increasing importance is the development of non-invasive tools to image the heart.
Researchers in the Whiting School of Engineering's Department of Computer Science have developed new motion tracking algorithms that can map out magnetic resonance images of cardiac structures in a way that improves the speed, quality and contrast of cardiovascular magnetic resonance imaging (MRI).
MRI is a widely used means for screening the heart and surrounding blood vessels. MRI uses radio waves to generate energy and then measure the response or signal of a specific tissue in the body. It uses this data to build a 3-D image of the internal organ being studied.
The process for creating images through an MRI requires obtaining data in small increments and then reconstructing a 3-D model by performing complicated mathematical manipulations on the data. The process for obtaining high-resolution images is very slow and time-consuming.
For the cardiovascular system, the data acquisition occurs over multiple heartbeats, causing the coronary arteries and valves being studied to move substantially while the image is taken. This often introduces artifacts and produces blurred, low-contrast images of the heart.
Additionally there is great variation between the respiratory functions of different patients, further complicating physicians' efforts to analyze and predict the motion of the heart and correct for it through motion compensation algorithms.
The Computational Interaction and Robots Lab, led by computer scientist Gregory Hager, designed a novel computer program called DEMOTRACS that is used to simplify MRI of the heart in real-time. A report on the computer program appears in the June 2007 proceedings of the IEEE Conference on Computer Vision and Pattern Recognition.
By tracking the arteries associated with the heart in low-resolution magnetic resonance images at high speeds and then "subtracting" the artifacts that are out of place, the new technique produces images that show the heart as if it were immobile.
After an initial scan of the heart, the captured magnetic resonance images are then used to re-position the next scan of the heart in a specified plane depending on whether the heart valve or arteries are being imaged. After continuous imaging, the algorithm then extracts the motion of the surrounding tissue based on the fixed plane of the image.
By tracking the heart structures across a host of resolutions, from temporal and spatial resolutions to high and low resolutions, a final image free of the artifacts caused by heart motion can be generated.
The algorithm can then be adapted by creating a template, or a stored reference window, for each specific region of the heart, thereby producing highly accurate estimates of the motion of each structure and the necessary extraction to produce an accurate image. Multiple templates were developed to account for the various deformations that occur as the heart beats, optimized by a standard coordinate mapping system.
So far the multiple-template algorithm developed by the Hopkins team has been able to image the heart and various cardiac structures accurately in magnetic resonance images in a variety of cardiac cycles. This development has been shown to greatly increase both the speed and quality at which MRI resultsare produced.
The researchers are now working towards integrating their algorithm with an MRI scanner to evaluate their motion-compensation techniques on human volunteers. Their estimation techniques will then be optimized to assist physicians in making accurate and early diagnoses of heart disease in at-risk patients.
