In this section, we summarize some of the salient aspects of validating electrocardiographic inverse solutions and identify what we consider the most pressing requirements for continued progress.
A ubiquitous requirement of validation is to quantify the level of error between the estimated results and some ground truth or gold standard. For electrocardiographic inverse solutions, even assuming such a gold standard is available, however, there does not yet exist a suitable metric with which to express this error. Statistical parameters such as correlation, relative error, and root-mean-squared error are inadequate because they are measures with no link to the electrophysiology or pathophysiology that electrocardiographers seeks to detect. For example, the presence of a small, isolated maximum or minimum in a measured isopotential map can suggest underlying abnormalities yet barely affect the overall statistical error when compared to a computed estimate that does not contain this feature. More generally, experienced human readers can detect the presence of specific features in distributions that one cannot capture in a robust automatic metric. As a result, any comparisons between estimated and measured results in an electrocardiographic validation study must be checked manually and often described qualitatively.
There is often a virtually unlimited number of different tests that are possible in a validation study. As a result, it is necessary to develop appropriate validation strategies, i.e., to extract from the infinite range of possible variations a reproducible subset of relevant features, parameters, and results. Such a strategy should include identification of the relevant parameters that are available for adjustment within the validation scheme. For example, in an experimental validation, a list of such parameters might include pacing sequence, heart rate, overall state of the heart, and choice of species. One must also decide how to impose shifts and random noise errors on the geometric model or conductivity values. At present, there exist no uniform guidelines for creating such a validation strategy. Defining such focused testing procedures will not only streamline execution and reporting, but also allow comparisons between methods proposed by different investigators.
A far greater hurdle to comparisons between investigators and laboratories is the lack of commonly available datasets. The measurements from the isolated heart and torso tank preparation first described by Taccardi, Colli-Franzone, et al. serve as the rare exception and have been used by several other groups around the world. If robust and comparable validation is to occur, it is necessary to collect many more such datasets and provide these to the research community in documented, electronic form. In recognition of this need, a goal of the recently established NIH/NCRR Center for Geometric Modeling, Simulation and Visualization for Bioelectric Field Problems at the University of Utah (www.sci.utah.edu/ncrr) is to provide such a service. Not only will isolated heart/torso tank data from within the Center be made available, but other investigators will have the opportunity to submit their data to the pool.
There are many reasons for optimism in the field of electrocardiographic inverse problems. The return to whole animal measurements by the Hunter group in Auckland and Oxford Universities, with the assistance of many new and powerful technologies for imaging and acquiring electrical data, offers new potential for highly realistic validation studies in complex geometry. The recent clinical studies that Tilg and the groups at the Technical University of Graz and the University of California at San Francisco have recently conducted represent the first example of a true patient based validation. These studies were also made possible by improvements in technology, in this case catheters that measure simultaneously activation time and spatial location. This convergence of technology and collaboration among engineers, experimentalists, and physicians will certainly result in new discoveries in the near future based on improved validation.
A further encouraging development in inverse electrocardiography is the recent interest of industry in all aspects of mapping, both cardiac and body surface. The CARTO system for endocardial mapping, the implementation by Endocardial Solutions Inc. of a non-contact endocardial mapping by means of an inverse solution, and the appearance of a clinical body surface mapping system from Meridian Medical Technologies all indicate that industry recognizes the promise of inverse solutions in medicine.
With continued effort and careful validation using some of the methods
described here, there can be little doubt that progress in inverse
solution research and application will continue to accelerate.