The application of a priori structural information based regularization in image reconstruction in magnetic induction tomography

Abstract

Magnetic induction tomography (MIT) is a non-invasive contactless modality that could be capable of imaging the conductivity distribution of biological tissues. In this paper we consider the possibility of using absolute MIT voltage measurements for monitoring the progress of a peripheral hemorrhagic stroke in a human brain. The pathology is modelled as a local blood accumulation in the white matter. The solution of the MIT inverse problem is nonlinear and ill-posed and hence requires the use of a regularisation method. In this paper, we describe the construction and present the performance of a regularisation matrix based on a priori structural information of the head tissues obtained from a very recent MRI scan. The method takes the MRI scan as an initial state of the stroke and constructs a learning set containing the possible conductivity distributions of the current state of the stroke. This data is used to calculate an approximation of the covariance matrix and then a subspace is constructed using principal component analysis (PCA). It is shown by simulations the method is capable of producing a representative reconstruction of a stroke compared to smoothing Tikhonov regularization in a simplified model of the head.