A cardiac potential mapping system

A Cardiac Potential Mapping System

Cecil W . T h o m a s , P h D , * K e n n e t h L a u r i t a , MS,-I- G a n g S u n , PhD,:I: J e r o m e L i e b m a n , M D , w a n d A1 L. W a l d o , M_D-t-

A b s t r a c t : The technical aspects of a multiple-purpose cardiac m a p p i n g system are presented. The authors begin with a brief history of hardware and software

development and then concentrate on the major problems in acquiring highquality recordings from the torso surface or from the epicardial surface and on the processing of the signals for display of color maps or other analysis. To achieve the desired adaptability to a variety of cardiac applications and experiments, they incorporated three parallel microprocessors that can record the signals from 240 electrodes and simultaneously provide display and analysis of the incoming cardiac data. While the parallel processors and modular software offer computational and flexibility advantages, the user interacts with an ordinary AT-compatible computer. This discussion is not a survey of the systems used in the many research laboratories or specific cardiac applications but is focused on experience in developing the specific hardware for such a multipurpose instrument, and less specifically on the software that makes the system easy to use and adaptable to a vareity of experimental and clinical situations.

Cardiac potential mapping has changed during the past several years, as the early experimental work in research laboratories has moved increasingly into clinical laboratories. The early work in body surface potential mapping (BSPM) in subjects and patients was paralleled by experimental epicardial mapping and clinical electrophysiological studies. The major objective of cardiac mapping is a description of the two-dimensional or three-dimensional distribution

of potentials in the heart. Even the activation maps or isochrones represent the spatial (and temporal) distribution of potentials. Thus, the c o m m o n element in a whole range of cardiac evaluation is the acquisition and visualization of the potentials. Such tests require simultaneous recording from m a n y electrodes, some machine processing of the acquired signals, and a display for h u m a n visualization and evaluation. Since the volume of data is quite large, the cardiac information must be efficiently presented, which requires some machine processing (if not analysis) of the recorded signals prior to display. Several research groups have developed cardiac mapping systems to address one or more aspects of cardiac potential visualization and analysis. Each group's data acquisition and display methods are skewed by the needs of the research projects, the interests of the investigators, and available funding. Ours is no exception. Rather than try to describe some optimum system, even if one exists, we will briefly describe our experiences with the technical

*From the Biomedical Engineering Department, Case Western Reserve University, Cleveland, Ohio. "]-From the Department of Medicine, Case Western Reserve University, Cleveland, Ohio. ~From Imaging Research Inc., Brock University, St. Catherines, Ontario, Canada. w the Department of Pediatrics, Case Western Reserve University, Cleveland, Ohio. Supported by many years of grant support from NIHLB (most recently Grants HL-17931-11 and HL-33343-02) and a Research Initiative Award from the American Heart Association. Reprint requests: Cecil W. Thomas, PhD, Biomedical Engineering Department, Case Western Reserve University, Cleveland, OH 44106.

64

Cardiac Potential Mapping

(hardware and software) aspects of cardiac potential mapping. In the spirit of the ISCE meeting, we will present opinions along with a report of our experiences and observations.

Brief History Our first-version system was assembled about 13 years ago for body surface potential mapping in pediatric cardiology.] Its major limitations were noise from the analog tape recording and the excessive t u r n a r o u n d time in using hardware in different locations. The analog tape provided ample recording time for several patients, but the noise level exceeded 60 ~,V rms. The tape was physically carried across the street from the clinical laboratory to an engineering laboratory, where the data were played back for digitization using a PDP 11/45 minicomputer. The maps were constructed and photographed for study by the whole research group. In spite of its shortcomings, that "system 1" served as a research tool while we gained experience in data collection, signal processing, imaging, and map interpretation. Recognizing the limitations of that system, we began to construct a self-contained stand-alone system 2. Digital recording replaced the noisy analog recording, a microcomputer replaced the general-purpose minicomputer, and floppy disk storage replaced the analog tape as the m e d i u m for archiving. That system also had some inadequacies, so we stopped short of a totally stand-alone system; we did not implement the imaging software on the system's Intel 8086 processor. Instead, the basic software was transferred from the PDP 11/45 (used for System i) to an IBM PC-AT, and map generation continues. Thus, the second system uses an 8086-based data acquisition system and an AT-based computer for creating maps. x'3 While this physical arrangement is not optimal, it does offer the opportunity to implement and experiment with software in a clinical setting and to use surface maps in various clinical studies. In fact, system 2 has been in continuous operation in pediatric cardiology for about 6 years. 4-1s During our early use of system 2, the emphasis in m a n y research laboratories has shifted toward epicardial and endocardial potentials and activation maps. While our major focus had been on the issues in pediatric cardiology and mapping of torso surface potentials, research expanded to adult cardiology with more left heart problems. We had learned to use surface potential mapping for spatial localization, which could be applied to infarcts, ischemia, and arrhythmias. The motivation for direct epicardial

9 Thomas etal.

65

mapping was based on the view that the epicardial potentials are closer to the source,19 and thus provide new diagnostic data that could be used in conjunction with existing tests, or perhaps reduce the need for some tests. Based on the extended application of potential mapping, we halted the development of system 2, leaving it as a two-part system for surface potential mapping. Development of a new system, system 3, was initiated to meet the combined demands of torso surface, epicardial, and endocardial mapping. The overall design was naturally based on our experience with the early two systems for torso body surface potential mapping, but the new system needed to be adaptable to different cardiac applications, and safe for use in a catheterization laboratory or operating room. We also required that the system be completely self-contained in a single unit for data acquisition, processing, and map generation, while providing for visual and automatic analysis of potential maps or activation maps, or both.

System Characteristics and Variations Electrodes The electrode configurations vary significantly among research groups and with different applications. While epicardial mapping can be done with one probe, and two electrodes can detect a surface ECG, most mapping systems have a m i n i m u m of 32 electrodes. The upper limit is based on both economic and technical reasons. Larger arrays require more hardware in each c o m p o n e n t of a mapping system, and thus the cost is increased. Technical limitations include field distortion caused by a very large array of closely spaced electrodes and the problem of conducting the signals from the electrodes to the electronics. Several groups have settled on 150-240 electrodes for BSPM and epicardial recordings. The appropriate n u m b e r surely varies with the objectives of the measurement. If one wishes to see general characteristics of surface maps, a small n u m b e r of electrodes may produce maps with acceptable error. If one is interested in the inverse problem or spatial regions (torso or epicardial) where the potential gradients are large, the n u m b e r of electrodes must be significantly larger. Aside from numbers, the type of electrode is far from standard. Motion at the skinelectrode interface is a constant consideration, if not a problem. Wet electrodes on the torso surface rain-

66

Journal of Electrocardiology Vol. 22 Supplement

imize the contact problem, but the conducting gels can produce electrical paths among electrodes. On the epicardial surface, the contact is naturally more moist but the motion is more severe. In our early research, we decided to use dry electrodes on the torso surface to avoid the problem of field distortion.X A total of 180 electrodes was placed in a vest for fast application of all electrodes simultaneously. Certainly, speed and convenience were achieved; the donning and fit adjustment can be completed in 2 minutes or less. Good contact can be achieved in all but about 1-3% (typically) of the electrodes, where the problems are usually in bony regions such as the spinal or clavicle regions. The dry electrodes present higher electrical impedance, but because current instrumentation amplifers have much higher input impedances loading effects are not a problem. For epicardial recording, the ventricular sock is commonly used for monitoring the whole ventricle. The principle is similar to that of the surface vest: fast application of a flexible structure for holding the electrodes in place. For studies requiring higher spatial resolution, such as arrhythmia studies, a surface patch can be used to apply a large number of electrodes to a small region of the atrium or ventricle. We use 192 electrodes to map the activation in regions of reentry phenomena. The whole aspect of electrode design and electrode application currently has adequate solutions, but probably no one would claim to have an optimum solution. Electrode research continues, along with the development of biosensors; their electrical properties and sizes will improve. The electrode considerations are not purely hardware: much of the software processing in a mapping system is devoted to processing, evaluation, and even editing to guarantee the quality of the recorded signals. The primary determinant of data quality remains electrode contact.

Amplifiers The signal from each electrode must be conditioned for impedance buffering, amplification, bandwidth limitation, and interference rejection. For unipolar recording, each amplifier receives a signal from one electrode and a reference such as the Wilson Central Terminal for torso surface recording. 3 For bipolar recording, two electrodes are connected to each amplifier. In system 3, we incorporated variable gains (i00, 200, 1,000, 2,000) and variable lower cutoff frequency (0.05 and 1.0 Hz). x~ These are selected from the operator keyboard; a controller

switches to the resistor configuration appropriate for the desired cutoff frequency and gain. Each amplifier has a first-stage instrumentation amplifier with lasertrimmed resistance to maximize the common mode rejection ratio (CMRR). The second and third stages adjust the gain and bandwidth. The amplifiers are arranged with 16 per board and up to 15 boards. The boards also contain sample and hold circuits, analogto-digital conversion, and multiplexing. The amplifier portion is off-the-shelf technology, and except for leakage current considerations, the I6-channel boards are standard electronics technology.

Sampling At each sampling instant, all sample-and-hold circuits are activated simultaneously. Then the outputs of the sample-and-hold circuits are sampled by a multiplexer-ADC (analog-to-digital coverter). This provides effectively simultaneous sampling of all signals, thus avoiding the problem of time skew among signals. The analog multiplexer combines the 16 signals on a board, so that a single ADC is required for each board.

Noise In any data acquisition system for low-level signals, a major measure of quality is the noise level. Sources of system noise include quantization in the analog-to-digital coverter, sampling skew when sample-and-hold circuits are not used, electronic noise, transient or sustained pickup from other instruments, and interference from power lines. In some environments, radio and television signals present troublesome interference. With careful design techniques, all of these noise components can be controlled. 2~ The electronic noise is limited by careful component selection and board layout. A major design effort focused on minimizing the powerline interference. The CMRR must be carefully maximized at the component level and in the selection of cables. The leads to electrodes and reference must have closely matched resistance and capacitance, so that impedance mismatch does not introduce powerline interference into the differential mode signal and lower the real CMRR. Also to reduce common mode interference, the shields of the signal cables are driven with the common mode voltage from the instrumentation amplifiers. The powerline interference can also be reduced by providing a patient ground at the right leg. Since a direct

Cardiac Potential Mapping

ground is not safe, a driven right leg circuit or a virtual ground can be used to provide the ground potential without also providing a significant current path to ground. We chose a virtual ground with a 470 kf~ current limiting resistance. By these straightforward methods and simple considerations, such as keeping the patient away from Power lines, the comm o n m o d e 60-Hz voltage can be limited to 100 ~V or less. The differential mode 60-Hz c o m p o n e n t is t h e n negligible. Our tests show that we can limit the total noise to 10 izV rms and usually 5 - 7 g,V rms for cardiac applications. These measurements were m a d e digitally in the T-P region of the ECG signals recorded from a resting normal subject. The signals were digitized, time segments were manually selected, and the rms values were calculated automatically. With the inputs grounded, the same recordings give 0 V rms; the noise is m u c h less than one quantization level. Our quantization noise is determined by the choice of a 12-bit analog to digital converter. With a maxi m u m expected cardiac potential of 10 mV, the quantization step is 4.883 p,V. Since most patients have m a x i m u m potentials in the 2 - 5 mV range, baseline drift does not cause signal saturation. For a given ECG signal, an 8-bit ADC may be sufficient, but the 12-bit ADC offers lower quantization noise, some margin for baseline drift, and better definition of lowlevel potentials. We looked briefly at the possibility of eliminating the sample-and-hold circuits to save on system p o w e r consumption. However, without the sampleand-hold circuits, the signals are no longer sampled simultaneously, and sampling skew becomes a consideration because the difference in timing introduces an error in voltage. If we require that the error due to sampling skew be no larger than the quantization error (4.883 ~V), the m a x i m u m skew is about 10 ~sec. In our system, we sample in groups of 16 signals, which would introduce a skew of 100 p,sec or more. With different multiplexing, the skew can be reduced, but the larger number of analog-todigital converters offsets the savings. Therefore, we decided to retain a sample-and-hold circuit for each signal. Sample-and-hold circuits are designed for speed, but the circuits use more current to charge the "hold capacitor" more rapidly. Since we do not need such speed in our multiplexing scheme, we opted to design our o w n sample-and-hold circuit with the appropriate compromise between speed and power requirement. That work led to a fourfold reduction in current requirement for each sample-and-hold. 21 The significance of this reduction goes beyond the

9 Thomas et al.

67

analog circuits: it indirectly reduces leakage current, as described below.

Leakage Current An underlying motivation in most of our design decisions was the requirement for limiting the possibility of microshock by limiting current pathways to ground. The various considerations include circuit design, c o m p o n e n t selection, board layout, grounding configurations, system power isolation, isolating analog and digital grounds, and isolation coupling of the digital signals. The goal is to limit leakage current to 10 ~A during any failure scenario. For example, suppose the patient is in a motorized metal bed with an electrical fault which places 120 vac on the metal bed. If the patient touches the bed, our system must not allow more than 10 ~A of current to ground. Similarly, if the patient touches the chassis of another instrument, we must limit the path to ground. This is not a difficult task for torso surface mapping, where microshock is less likely, but our goal is to have an instrument that can also be used in the catheterization laboratory and operating room. Therefore, limiting the leakage current is a vital issue and a difficult design problem. In system 2, the current to ground is monitored and an alarm sounds w h e n e v e r leakage current exceeds 10 ~A, and the patient is automatically disconnected from the machine. System 3 can meet the leakage current requirements.

Defibrillation Protection In the cath laboratory and operating rooms, it is possible that the patient will need defibrillation. If a system is connected to the patient w h e n defibrillation voltages are applied, the system input must be able to tolerate the high voltage. While we have no direct experience in this problem, some reasonably straightforward techniques are available. Some protection circuits utilize devices called "spark gaps," which are gas discharge devices. The simplest and most c o m m o n protection is a network of neon bulbs and a spark gap that become active at high voltages, but that do not disturb the normal operation at the voltage level of cardiac potentials. These components essentially shunt the defibrillator current from the system electronics, so that the reversible breakdown of the protection circuit prevents the irreversible breakdown of the system components. The use of neon bulbs reduces the n u m b e r of the larger and

68 Journal of Electrocardiology Vol. 22 Supplement more expensive spark gap devices. Another factor is the capacitance of the spark gaps (and neon bulbs), because of the potential leakage current path through those devices. Although only about 1-5 pF each, this is excessive when a device is connected at each of 200 electrodes. A completely floating input could also tolerate the defibrillator voltages. This whole area is in the category o f " a r t " ; engineers seem to use something that works without knowing why. For example, the protection circuits must be physiologically separated from the inputs by some minimum distance. Given the use of our system to date and in the near future, we chose to use the manual/mechanical disconnect, thereby avoiding the additional leakage current introduced by the defibrillation protection circuits.

Storage Media For BSPM, analysis is most often confined to a single heart beat. Based on that fact, system 2 was initially designed to record for only 1.5 seconds, with DMA storage in memory (RAM). While this brief recording time may be sufficient for some BSPM analysis, longer record times are either necessary or desirable in the other cardiac applications. By adding more RAM, we could record for 15-16 seconds, but for longer record times we want to use a different media. Either disk or magnetic tape could extend the record time to several minutes using current technology. In studies of drug effects or cardiac arrhythmias, the longer time interval is a necessity. Currently, standard magnetic tape (two units) can provide continuous recording on nine-track tape, with tape reloading about every 5 minutes. Large hard disks can record for longer times, but they must be backed up. Streaming tape (8-mm) can be used, but maintaining the streaming rate to avoid startstop can be tricky. Parallel cartridge disks would work, but replacing the cartridges would be too timeconsuming. Of course, one can always record on analog tape. The most attracting device is the optical drive with reusable 650-MB cartridges. The sustained transfer rate is still marginal for our maximum number of electrodes and maximum sampling rate, but that technology is improving, and the cost is quite reasonable.

Computer We selected the IBM PC-AT as the computing device because of its cost, availability and flexibility in

hardware and software. It has the DMA capability, sufficient graphics with software, and computational speed to handle most algorithms. The PC has an open architecture that allows us to add boards for specific tasks. We added a Sritek board with 68020 processor to handle the data acquisition and processing while the AT's 80286 processor remains free for processing, analysis, or display. A video card (from Number Nine) provides display and has a graphics processor. The three processors (Sritek 68020, IBM 80286, and Number Nine graphics processor) can operate in parallel. For example, during data acquisition at the maximum rate, the 68020 manages the acquisition and is about 80% free for additional processing, the 80286 can access the incoming data for processing, and the graphics processor can create the video display of potential signals, isochrones, etc. The video processor software can display 8-bit color mapfi of potentials or isochrones and equipotential contours or isochrones. Furthermore, one-dimensional signals can be displayed for checking the recorded data, generating 12-1ead ECG signals, or producing other signal tracings. The operator interacts with the 80286 as if it were the only processor in the AT, but the software can be quite flexible in taking advantage of the parallel processors.

Data Collection and Monitoring Software When a large number of signals are being recorded from some anatomical surface, the operator-technician needs some feedback on the quality of the signals, the integrity of the system, and an indication of problems. For a small number of electrodes recorded for a short time (eg, in vectorcardiography), this feedback is not so important. However, when we record from 200 or more electrodes, the technician needs help in determining data quality. Our approach is to have the software lead the technician through a short checklist to select the desired gain and bandwidth settings, enter patient identification, identify electrode configuration, and identify any other routine data or options. Then, when the electrodes are in place and data acquisition begins, a video display helps the technician determine whether everything is proceeding normally. One mode of recording uses the storage device as a circular buffer (ie, recording and re-recording on the same media until the technician interrupts recording). This is especially useful in arrhythmia studies, where the investigator monitors the patient or animal until an abnormal beat occurs. Then by recording past the abnormal beat, the buffer contains data

Cardiac Potential Mapping

before and after the beat of interest. In drug studies, the same interrupt can be used to record data before and after administering a drug. Another mode is "gating," in which the data acquisition is synchronized with a stimulus, respiration, or other event.

Baseline Adjustment Many techniques are available for baseline adjustment or removing baseline drift in ECGs. Most of those techniques work well for individual signals, but not well enough for maps. This was a major surprise for most of us w h o thought that baseline adjustment was a simple straightforward problem. The difficulty is due to the fact that our maps show spatial distribution of potentials, while the baseline adjustment is performed on individual temporal signals. If the baseline removal in only one signal has a residual error at some point in time of 3%, the error will not be noticeable in a display of potential versus time. However, w h e n the signals are displayed in a spatial format, the error at the electrode site becomes very obvious. Spatial smoothing or contour smoothing makes it less obvious by spreading the error over a neighborhood. The usual solution is "linear trend" baseline adjustment, 22 assuming that the signals should be zero potential at the beginning and end of each heart cycle, and sometimes in the P-Q interval. Linear trend removal is adequate, but any residual errors appear as faults. It is not clear that a higher-order polynomial removal will be better (or even as good), but some use of baseline drift information in adjacent cycles may be useful. We chose to use the first-order adjustment and concentrate on the fault detection.

Faulty Signal Detection The other major task we call fault detection, which means identifying those signals that cannot be corrected by baseline adjustment. The fault could be due to poor electrode contact, wire breakage, or some other source; our system keeps a record of faults so that recurring faults can be examined for underlying problems. In our earlier system, the fault detection was based on empirically defined tests for excessive baseline offset, excessive mean-squared noise, and discontinuities in baseline (ie, temporal measures). Experience with that system showed that temporal measures are insufficient for detecting a majority of faults; purely temporal tests can detect less than half of the faults in BSPM. We conclude that precise fault

9 Thomas et al.

69

detection algorithm must include not only temporal measures, but also spatial tests (ie, comparing a signal from an electrode with the signals from neighboring electrodes). 22"23 The problem is exaggerated by the sensitivity of a system with large numbers of electrodes. If we had only 10 or 30 electrodes, a small error in a given electrode would be overlooked because virtually no spatial data would be available to help detect the error. However, with large numbers of electrodes, a small potential error at one spatial coordinate can be detected by simple comparisons with measured potentials at nearby coordinates. If the nearby potentials are not available, the error cannot be detected. Thus, the large n u m b e r of electrodes is simultaneously a luxury and a curse; we can better detect small errors in spatial potential gradients, but we must also be more precise in processing to reduce errors in potential.

Interpolation Another deceiving problem is the interpolation among the measured potential values on some surface. The interpolation is necessary to make the map sufficiently large and provide sufficient definition of the potential gradients. If the surface were a uniform grid with potential values at each grid intersection, the solution would be simple; any of several algorithms could be used to interpolate in two dimensions. However, in general the electrodes are not uniformly spaced, some regions of the anatomical surface may not have any electrodes, and some potential values may be missing because of faults. Bilinear interpolation can be used successfully. It is not clear that polynomial, spline, Hermit methods, or linear filtering approaches offer any advantage. We should point out that contour maps have the s a m e inherent interpolation problems. While the interpolated values are not retained or displayed, contour positions are computed on a discrete grid by interpolating among the original potential values.

Color Maps Several years ago, we decided to use color to represent the cardiac potentials in the maps. The primary motivations were simplicity (no data-dependent decisions), ease of viewing (regions of similar potential are more easily seen as regions), and speed (display requires no computation). At that time the inexpensive hardware could display only 15 colors (and black), so we used all 15 in both the negative

70

Journal of Electrocardiology Vol. 22 Supplement

and positive potential ranges. Thus, we used 31 levels, but at the expense of redundancy in the positive and negative scales. 2 Since that first system was built, new inexpensive hardware has become available with 4,096 or more colors. Given that the redundancy is a disadvantage (albeit minor), we decided to define a better color scale. We started by defining several different strategies for defining a color scale as a path through the three-dimensional color space. Since w e are interested in video display with red, green, and blue (RGB) primaries, we confined our attention to the subregion of color space accessible with the RGB color system. For RGB, the cones in the general bicone space become triangular pyramids in a space with coordinates of hue, saturation, and intensity. We also decided to maintain a sharp difference between low positive and low negative potentials (ie, a discontinuity at zero potential). This deviated from the almost continous scale proposed by other groups. In early work, we formulated a cardinal rule that adjacent colors must be similar, while being uniquely identfiable on the scale relating color to potential values. A collection of such rules led us to a relatively small number of approaches to defining the path through color space. Based on those rules, a n e w color scale evolved, 24 and it has been adopted for torso and epicardial potential maps.

Alternative Image Formats

To maintain compatibility with other laboratories, we developed software for drawing contours. We also considered the possibility of displaying "texture m a p s " using binary textures or patterns, as in the Apple Macintosh software. The definition of the appropriate textures seems similar to (but simpler than) the color scale selection. The main motivation for the texture or contour maps is not dissatisfaction with the color maps, but the enormous costs of publishing in color. M a n y joumals do not accept manuscripts with color figures; other assess enormous page charges. The contour and texture mapping were developed in response to economic considerations. The primary practical perceptual difference bet w e e n contour maps and color maps is that a h u m a n views contours as a set of curves, while the color maps are viewed as a set of regions. The curves "look better" w h e n they are smooth, thus contour smoothing is usually applied to maps of real data. However, for color maps, the perceptual focus on regions removes the necessity for spatial smoothing.

On-line Isochrones In epicardial mapping work, much of the important information can be displayed in the form of isochrones. In short-term studies or more lengthy experiments on arrhythmias or drug effects, the isochrone can be the major tool for monitoring the progress of the experiments. In m a n y cases, the isochrones contain all the necessary information. Because of the multiprocessor approach in our system, the isochrone display can parallel the recording of potentials. This dual-parallel capability is necessary in various studies where isochrone monitoring is the primary visual feedback during the experimental procedure and later off-line potential processing is also required for more detailed and more quantitative study.

Image Analysis A biopotential mapping systems can generate an enormous volume of data and images. The challenge is to use our computing power, take advantage of the superior h u m a n powers of visual perception w h e n appropriate, and maintain sufficient flexibility so that the system can adapt to multiple experimental protocols. Our general systems approach has a three-processor hardware configuration and a modular software structure. Thus the challenge translates into software algorithms that can utilize the system flexibility and meet the changing demands of the research and clinical cardiac laboratories.

Acknowledgments The authors thank a large n u m b e r of graduate students who contributed to the research and specific work described here.

References 1. Plonsey R, Ko W, Liebman J e t al: Body surface potential mapping project at CWRU: p. 77. Yamada K, Musha T, Harumi K (eds): Advances in body surface potential mapping. University of Nagoya Press, 1983 2. Kavuru M, Vesselle H, Thomas CW et al: An on-line color displayed body surface potential mapping system. Proc Int Congr Electrocardiogr, 1985

Cardiac Potential Mapping

3. Kavuru MS, Vesselle H, Thomas CW: Advances in body surface potential mapping (BSPM) instrumentation. In Liebman J, Plonsey R, Rudy Y (eds): Pediatric and fundamental electrocardiography. Martinus Nijhoff, 1987 4. Liebman J, Thomas CW, Rudy Y, Plonsey R: Electrocardiographic body surface potential maps of the QRS of normal children. J Electrocardiol 14:249, 1981 5. Liebman J, Thomas CW, Salamone J et al: Quantification of electrocardiographic body surface maps of the QRS and T waves of normal children. Jpn Heart J 23:409, 1982 6. Liebman J, Rudy Y, Diaz P et al: Electrocardiographic body surface maps in advanced right bundle branch block, p. 217. In Yamada TMK, Harnmi K (eds): Advances in body surface potential mapping, University of Nayoya Press, 1983 7. Liebman J, Thomas CW, Rudy Y e t al: Clinical data with a color displayed 180 electrode ECG-BSPM system. Proc Conf IEEE/EMBS. Columbus, Ohio, 1983 8. Liebman J, Rudy Y, Diaz PJ et al: Body surface potential mapping: partial right bundle branch block versus right ventricular hypertrophy with terminal right conduction delay. J Am Coil Cardiol 3:496, 1984 9. Liebman J, Rudy Y, Diaz P et al: The spectrum of right bundle branch block as manifested in electrocardiographic body surface potential maps. J Electrocardiol 17:329, 1984 10. Liebman J, Thomas CW, Vesselle Het al: Observations on ventricular hypertrophy utilizing body surface potential mapping. Proceedings of the World Congress in Pediatric Cardiology. New York, 1985 11. W i d m a n L, Liebman J, Thomas CW et al: Quantification of electrocardiographic body surface potential maps (BSPM) of the QRS and T of normal young adult males. Proc Int Congr Electrocardiol 1986 12. Liebman J, Thomas CW, Fraenkel R, Rudy Y: Preliminary observations of the RVH in body surface potential mapping (BSPM) of pulmonic stenosis (PS) and arterial septal defect (ASD). Proc Int Congr Electrocardiol 1986 13. Liebman J, Rudy Y, Thomas CW, Plonsey R: RVH with terminal right conduction delay versus partial right bundle branch block (utilizing body surface potential maps), p. 389. In Lieman J, Plonsey R, Rudy Y (eds): Pediatric and fundamental electrocardiography. Martinus Nijhoff, Amsterdam, 1987

9

Thomas et al.

71

14. Liebman J, Thomas CW, Salamone R et al: Electrocardiographic body surface potential maps of the QRS and T of normal children: qualitative description and selected quantitations, p. 381. In Liebman J, Plonsey R, Rudy Y (eds): Pediatric and fundamental electrocardiography. Martinus Nijhoff, Amsterdam, 1987 15. Cohen MH, Beder SD, Thomas CW, Liebman J: Characteristics of right ventricular activation detected by body surface potential mapping among post operative tetrology of Fallot patients with and without arrhythmia. Am J Cardiol 60:638, 1987 16. Widman L, Liebman J, Thomas CW et al: Electrocardiographic body surface potential maps of the QRS and T of normal young adults: qualitative description and selected quantifications. J Electrocardiol 21:121, 1988 17. Liebman J, Thomas CW, Rudy Y: Body surface potential mapping in conduction abnormalities (with particular reference to congenital heart disease). In Mirvis D (ed): Body surface potential mapping. Martinus Nijhoff, Amsterdam, 1988 18. Liebman J, Thomas CW, Fraenkel R, Rudy Y: Analysis of the hypoplastic right ventricle utilizing electrocardiographic body surface potential mapping (BSPM). J Electrocardiol 22:195, 1989 19. Rudy Y, Messinger-Rapport BJ: The inverse problem in electrocardiography: solutions in terms of epicardial potentials. CRC Crit Rev Biomed Eng 16:215, 1988 20. Laurita K, Thomas C, Kavuru M e t al: Data acquisition system for cardiac mapping. Proceedings of the Tenth InternationalConference of IEEE Engineering in Medicine and Biology Society, 1988 2 I. Laurita KR: Hardware for cardiac mapping. MS thesis. Case Western Reserve University, Cleveland, 1989 22. Thomas CW, Lee D: Methodology in constructing body surface potential maps. p. 329. In Liebman J, Plonsey R, Rudy Y (eds): Pediatric and fundamental electrocardiography. Martinus Nijhoff, Amsterdam, 1987 23. Lee D, Thomas CW, Rudy Y, Liebman J: A knowledge-based approach in generating body surface potential maps. Proc Int Congr Electrocardiol 1986 24. Huebner WP, Thomas CW, Rudy Y, Liebman J: An improved pseudocolor scale for the display of body surface potential maps. Proc Int Congr Electrocardiol 1986