Improving image analysis techniques for electron tomography of the 3D architecture of Plasmodium falciparum

© 2014 Dr. Mauro MaiorcaElectron Tomography is an established technique for producing detailed 3D image volumes necessary for investigating the structure and function of cellular components at nanoscale. However, the analysis of Electron Tomography volumes is still a challenge. For example, identifying biological features and segmenting structures often requires skilled manual intervention and is time consuming and error prone. Unfortunately, the sample-preserving low exposure acquisitions and the partly unpredictable behaviour of electron-specimen interactions lead to an overall low contrast in reconstructed volumes. Furthermore, highly electron dense features, such as gold nanoparticles, can introduce severe reconstruction artifacts. An Electron Tomography volume is obtained by solving the mathematical inverse problem of 3D reconstruction from an incomplete tilt projection stack, discontinuously acquired from a specimen. Under these sub-optimal conditions, the inverse problem of Electron Tomography reconstruction is considered severely ill-posed (in the sense of Hadamard), and its solution, obtained by the reconstruction algorithms, suffers from instability and non-uniqueness. As a consequence, Electron Tomography reconstruction algorithms cannot guarantee a linear relationship between the reconstructed image intensity values and the densities of the specimen under examination. This prejudices the overall reconstructed image quality, quantitative analyses, and post-reconstruction processing of Electron Tomography volumes. Therefore, the analysis of pre-reconstruction pathways for the reduction of the ill-posedness of the inverse problem of Electron Tomography reconstruction is a fundamental step for obtaining better quality Electron Tomography image volumes. In this thesis, we posit that the Electron Tomography ill-posedness can be reduced by introducing specific constraints in pre-reconstruction images, and we present the design, implementation, and application of a family of pre-reconstruction filters. The specific constraints introduced are based on nonlinear anisotropic diffusion, nonlinear isotropic diffusion, and acquisition geometry of tilt projections, and they provide (respectively) applications for image quality enhancement, gold-nanoparticle artefact reduction, and serial sections discontinuity reduction. To validate our approach we established a set of experimental paradigms based on objective comparative analyses between ground test (processed and unprocessed test data) and ground truth (control data) on both simulated and biological case studies. Our filters were also validated by applying them to the analysis of specific structural and functional features of the malaria parasite, P. falciparum