Reconstruction and validation techniques for MRI-based electrical properties tomography in the brain

Abstract

With Electrical Properties Tomography (EPT), tissue conductivity and permittivity can be non-invasively derived from MRI measurements. As these tissue properties depend on tissue structure and composition, knowledge of conductivity and permittivity can be valuable in several clinical applications such as oncology. However, several challenges in EPT reconstruction have prevented clinical adoption so far. For example, the reconstruction of electrical properties remains difficult because it involves solving complex partial differential equations using noisy data. Additionally, validation of the reconstructions is an issue because ground truth properties cannot be measured in vivo. This is especially problematic for trust in deep learning based methods. The work in this thesis focuses on the development of methods for reconstruction and validation to address these challenges. The first step to improve EPT reconstruction in this work is the generation of a database with simulated brain data for EPT reconstructions. These data were shared open access to provide both a large and easily available dataset for training of data driven methods. Building upon this, a hybrid deep learning and physics based reconstruction approach was developed to improve reconstruction quality, by combining the strengths of data driven and physics based reconstructions. To improve validation of EPT, a three dimensional multi-compartment brain phantom was developed. This phantom consists of a 3D printed shell, that can be filled with solutions of known electrical properties. This phantom provides the opportunity to test reconstructions with a phantom of more realistic and complex geometry. Finally, a method was developed that can be used to quickly check the consistency of reconstructed electrical properties with measured data, even in absence of knowledge of ground truth electrical properties. This can especially help to provide trust for the adoption of data-driven reconstruction methods. With these new techniques EPT reconstruction has become more accurate and trustworthy. This is an important step towards the future clinical applicability of the technique.