1568 IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 57, NO. 7, JULY 2010 Ranking the Influence of Tissue Conductivities on Forward-Calculated ECGs David U. J. Keller ∗ , Member, IEEE, Frank M. Weber, Member, IEEE, Gunnar Seemann, and Olaf D ¨ ossel, Member, IEEE Abstract—This paper examined the effects that different tis- sue conductivities had on forward-calculated ECGs. To this end, we ranked the influence of tissues by performing repetitive for- ward calculations while varying the respective tissue conductiv- ity. The torso model included all major anatomical structures like blood, lungs, fat, anisotropic skeletal muscle, intestine, liver, kid- neys, bone, cartilage, and spleen. Cardiac electrical sources were derived from realistic atrial and ventricular simulations. The con- ductivity rankings were based on one of two methods: First, we considered fixed percental conductivity changes to probe the sen- sitivity of the ECG regarding conductivity alterations. Second, we set conductivities to the reported minimum and maximum values to evaluate the effects of the existing conductivity uncertainties. The amplitudes of both atrial and ventricular ECGs were most sensitive for blood, skeletal muscle conductivity and anisotropy as well as for heart, fat, and lungs. If signal morphology was con- sidered, fat was more important whereas skeletal muscle was less important. When comparing atria and ventricles, the lungs had a larger effect on the atria yet the heart conductivity had a stronger impact on the ventricles. The effects of conductivity uncertainties were significant. Future studies dealing with electrocardiographic simulations should consider these effects. Index Terms—Conductivity uncertainties, electrocardiographic forward problem, inhomogeneity, model simplification. I. INTRODUCTION T HE ELECTROCARDIOGRAPHIC forward problem con- nects the electrical signals generated in the heart with the resulting potential distribution on the body surface. Apart from the location and amplitudes of the cardiac sources, the electric field in the human torso is mainly determined by the size and position of the internal organs and structures (also referred to as inhomogeneities). However, the various tissues, fluids, and structures are known to vary with respect to their conductivity and degree of anisotropy [1], [2]. Thus, their influence on the surface ECG is not intuitively predictable. Manuscript received November 24, 2009; revised February 9, 2010 and March 17, 2010; accepted March 17, 2010. Date of current version June 16, 2010. The work of D. U. J. Keller was supported by the German Research Foundation (DFG SE 1758/2-1). The work of F. M. Weber was supported by Philips Re- search, Hamburg. Asterisk indicates corresponding author. ∗ D. U. J. Keller is with the Institute of Biomedical Engineering, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany (e-mail: David.Keller@kit.edu). F. M. Weber, G. Seemann, and O. D¨ ossel are with the Institute of Biomedi- cal Engineering, Karlsruhe Institute of Technology, 76131 Karlsruhe, Germany (e-mail: frank.m.weber@kit.edu; gunnar.seemann@kit.edu; olaf.doessel@ibt. uni-karlsruhe.de). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TBME.2010.2046485 Simulation studies have tried to characterize the influence of inhomogeneities on the computed body surface potential maps (BSPMs) based on dipole sources [3] and measured epicardial signals [4]. However, these studies do not address the lack of consensus in the literature about measured tissue conductiv- ity values [5]: The conductivities of all major tissues in the thorax differ by a factor of 2.3–16.5 between different studies (e.g., compare conductivity of the kidneys with values between: 0.0544 S/m [6] and 0.9 S/m [7]). One reason for these differences are the technological chal- lenges that arise when measuring conductivities in the low- frequency range required for electrocardiographic calculations. Other deviations result from different measurement techniques [8], measurements on different species, or even sample vari- ations within the same species. Furthermore, it is known that tissue conductivities change ex vivo after the sample has been excised [9], [10]. Finally, there are pathological conditions that cause changes in tissue or fluid conductivity [11], [12]. All this also influences simulated ECGs. The higher the uncertainty in the conductivity for a specific tissue, the more severe are the modeling errors it can introduce. Another important aspect in electrocardiographic simulations is the model complexity. In recent years, patient-specific models have become increasingly important. The 3-D imaging modal- ities like computed tomography (CT) or MRI produce high- resolution datasets of a patient’s anatomy. However, the subse- quent steps which are necessary for the generation of a compu- tational model remain time-consuming and labor-intensive as a complete automatic segmentation is rarely available. Hence, the question regarding the anatomical detail that has to be incorpo- rated into the patient’s torso model needs to be addressed. Thus, the aim of our paper is twofold: At first, we want to rank tissue and fluid conductivities according to their importance in two different ways. On one hand, we considered fixed percental changes of the conductivities to probe the BSPM’s sensitivity regarding conductivity alterations. On the other hand, we set conductivities to the reported minimum and maximum values. Hereby, we evaluated the BSPM’s variation due to the present conductivity uncertainties. In addition to that, we propose rec- ommendations that are targeted on facilitating the creation of patient-specific models. These recommendations will establish a link between torso model simplifications and the magnitude of corresponding modeling errors that are thereby introduced. In contrast to the previous studies, our investigations are based on current sources derived from a realistic cardiac simulation of a complete heart beat. This makes the results especially rel- evant to applications regarding possible clinical applications of cardiac simulations. 0018-9294/$26.00 © 2010 IEEE