Conclusion

This study provides quantitative data on the effects of including or excluding selected inhomogeneities and anisotropies when solving the forward ECG problem within a realistic finite element model of the human thorax.

The influence of these findings on the design of torso models for forward computations can be summarized in the following points:

1. Models consisting of a single inhomogeneous region will always incur significant errors compared to those that incorporate a more complete representation of torso inhomogeneity. No realistic single-inhomogeneity model was capable of getting within 10% (RE) of the fully inhomogeneous model.
2. The effects demonstrated by either adding or removing inhomogeneities appeared to have only minor influence on the location of features of the BSPM but instead altered the shape and magnitude of these features.
3. In a multi-component model, removal of subcutaneous fat, the anisotropic skeletal muscle, or the lungs from an otherwise ``complete'' model will result in errors comparable to the best of the single-inhomogeneity models. Hence, these components must be considered crucial to such a model of the thorax.
4. The influence of skeletal muscle depends strongly and monotonically on the degree of anisotropy incorporated.
5. The influence of different inhomogeneities depends on the potential field distribution in ways that are not easily predictable, both in terms of absolute errors and the relative errors induced by addition or removal of individual inhomogeneities.

The information provided in these results demonstrates the effects of various structures that may be included in finite element models of electrocardiographic fields. The actual value at which an effect becomes important is dependent on the particular modeling application. Details that change the results by a small amount may not be worth including for some modeling applications, while this same detail may be critical in a different application, such as solving the electrocardiographic inverse problem. The results of this study will allow researchers to make quantitative comparisons of different modeling options and help them better optimize the use of their computational resources.