Parametric EIT inversion with sparse model sampling

Abstract In constrained inverse problems, such as biomedicine where human anatomy constitutes a known prior, a parameterized model can at times be just as effective at producing an accurate image directly from EIT data without resorting to an intermediate conductivity map. A reduced measurement set of high-value data can be mapped directly to the limited number of model space parameters of interest. To achieve this direct inversion, the high-value data can be identified with the sensitivity volume method, and a sparse data sampling of the model space can be continuously interpolated to span the complete model space. A simple nearest-neighbor query in data space then maps directly to a parametric representation for reconstruction. This method is experimentally demonstrated in a salt water tub with an insulating cylindrical inclusion whose diameter and angular position constitute a two-dimensional model space. The parametric method demonstrated here requires an order of magnitude fewer data measurements than standard EIT to achieve equivalent accuracy within the parametric representation of the model, and can be expanded to 3D and systems of higher complexity such as organs within the human body to achieve high fidelity parametric reconstructions with minimal data collection and fast, real-time image generation and diagnostics.