The Echocardiographic Research Laboratory:
Mission and Methods

Edward A. Geiser has been the director of The Echocardiographic Research Laboratory (ERL) since its founding in 1978. The research interest of the laboratory focused on the technology of cardiac ultrasound, which has become widely accepted because it provides an effective method for real-time monitoring and evaluation of the patient's cardiac health. Additional reasons for its acceptance include its safety to the patient, its portability, and its relatively low cost.

While most echocardiographers typically make subjective evaluations of the cardiac health of the patient subjectively or make manual estimates of heart size using calipers held up to a video monitor, the mission of the ERL laboratory has always been to develop computer based methods for making measurements that provide rapid, accurate, and objective approximations of the epicardial and endocardial borders of the left ventricle (LV) for a sequence images acquired during the cardiac cycle. The reason for interest in this goal is that once these borders have been obtained, computations of frequently considered indices of cardiac health are immediate. These indices include chamber diameter (LVID), chamber area, wall thickness, and the fractional area change (FAC). If the 2-dimensional epicardial and endocardial surfaces in a 3-dimensional image sequence could be similarly approximated, then stroke volume, muscle mass, and the ejection fraction (EF) would also be available.

The first computer based methods developed at the ERL laboratory were semi-automated in the sense that the cardiologist was required to input estimates of the endocardial border at both end diastole (ED) end systole (ES). The computer based method would then compute the borders for all the intermediate frames as well as readjust the expert's estimates at ED and ES. This method was coded into a FORTRAN software package in 1981 and included in the Quantic 1200, an ultrasound image analysis device manufactured by Bruce Franklin, Inc., a then existing company. This semi-automated method was later expanded to detect both the epicardial and endocardial borders and led to a second generation analysis system, the Insight 2000. While this software performed admirably, it failed to be accepted for two fundamental reasons. The first was that it took several minutes (as well as some skill on the part of the user) to input the required manual ED and ES borders. The second reason was that the computer hardware (an 80086 microprocessing chip) was not capable of making the computations rapidly enough. In fact, computation times were typically about 4-8 minutes. Since the demands made on a physicians time is always an important factor in health care, the method failed to gain acceptance. Since the staggering advances made in the area of computer hardware during the past decade have reduced computation times from minutes to seconds, the current focus of the ERL laboratory is to develop fully automated methods that require little or no user input.

Methods

Since the mission of the ERL laboratory is in the area of automatic measurements or automatic diagnosis, other areas of imaging such as formation, visualization, and low-level image processing (eg, denoising and enhancement techniques) are of interest only if they aid in the goal of improving estimates. Because of the dual problems of noise and dropout in addition to the enormous patient variability typically present in echocardiographic images we have moved away from the ``bottom up strategies" (using mean filters and edge-detection techniques) found in the classical image processing literature, to a more ``top-down" strategy. For images acquired off-line, the first step in this strategy is to identify the sector scan of the image so that the patient data and EKG image information are avoided. (For images acquired on-line, the location of the sector is known.) Once the sector scan has been found, the next step is to identify the epicardial/pericardial interface along the posterior wall of the myocardium. The reason for this decision is that this interface is typically the most prominent feature in the image and provides an approximation of the center of the LV. From this center point estimates can then be made of the borders in 64 equiangular directions. The decision making process can vary considerable from one direction to another. Since tracking of the motion of a structure through time is a fundamental component in the interpretation of a patient study, estimates are always made using all the image frames acquired during the cardiac cycle from ED to ES. The computer based algorithm currently used in the ERL laboratory takes approximately 8 seconds when run on an IBM RISC/6000 workstation, which was purchased in 1991. When the same algorithm was run on a state of the art Silicon Graphics workstation, the execution time dropped to 1-2 seconds.

Testing

Since the appearance of an echocardiographic image depends on the acquisition device, the experience and skill of the technologist, the patient's anatomy and disease state, testing is a fundamental concern of the ERL laboratory. At the present time several thousand image sequences are archived on disk drives. To test our methods, various sequences were selected for inclusion in a test database. Included in this database are 110 image sequences selected as representative of good-excellent quality, 92 sequences were selected as of good quality, 100 were obtained consecutively, and 150 were obtained in conjunction with animal studies. For each of these sequences an echocardiographer used mouse-driven software to trace his/her estimates of the epicardial and endocardial borders at ED and ES. For the set of 110 good-excellent quality images, three different experts each made border estimates. Thus, not only can comparisons be made between the experts and the computer-based, but between the various experts as well. In this way, not only can updated changes to the computer-based methods be compared daily with a particular expert, but these differences can be compared with the variability between experts. Except for patients with technically inadequate image information and those with certain abnormalities (eg, a large pericardial effusion), the performance of the computer based methods is close to the performance of an expert.

Education and Teaching

The laboratory has been involved with interdisciplinary education for the past 15 years. In particular, graduate and post doctoral students from the areas of electrical engineering, computer science, and mathematics have all worked with us for periods of time ranging from 3 months in the summer to several years. Most of these have been supported by funds appropriated from our contracts and grants. In the last 3 years two students in computer science have written theses on automated methods for isolating structure in echocardiographic images in partial fulfillment of their requirements for a Masters of Science Degree in computer science. One student in the mathematics department plans to begin work in 1996 on her PhD dissertation on the problem of tracking motion in echocardiographic images. In the past 5 years, a number of 3rd and 4th year cardiology fellows have completed one and two month rotations with us. While graduate education is the main concern of our laboratory, we have have even cooperated with the Center for Precollegiate Education and Training to sponsor a high school student to visit for 10 weeks during the summer of 1994