Advanced methods for volumetric interpolation and multimodal fusion of tomographic sequences with application to 3D reconstruction in biomedicine

Recently, anatomical 3D models, thanks also to the increased diffusion of 3D printing technologies, have been successfully introduced in the clinical field as innovative tools to support the medical team in several tasks. Advances in 3D technology now provide a realistic representation of complex anatomies that can be used as an aid for diagnosis, surgical planning and training. In general, 3D reconstructions are more accurate when they are the outcome of the processing of high-resolution image sequences. On the other hand, if a sequence is characterized by few tomographic sections, only a coarse reconstruction is achievable and the artifacts that are produced impair the usefulness as a tool to support the physician. The study carried out during this thesis work has two objectives. The first is to provide axial interpolation methods that, applied to low-resolution sequences, allow a refined 3D reconstruction of the anatomical model to be obtained. The methods are inspired by compensated frame interpolation techniques (MCFI) developed for video processing applications and produce an estimate of the displacement vector field (DVF). The DVF is then processed and used to estimate intermediate ones. The performance of the proposed methods has been quantitatively assessed using sequences with simulated axial low resolution. The experimental results show that the proposed methods allow an effective sliced interpolation and the 3D models obtained clearly benefit from the increased axial resolution. The second objective is studying image fusion methods specific for biomedical images obtained from different acquisition methods, also termed as multimodal image fusion. More specifically, in this thesis magnetic resonance imaging (MRI) and magnetic resonance elastography (MRE) sequences have been taken into account. The objective of the processing is achieving a fused image containing the structural information from the MRI and the mechanical (tissue stiffness) information from the MRE. The proposed method is based on multiresolution analysis (MRA). The fused MRE image is obtained by adding the geometric details, extracted from the MRI, after being modulated by a suitable injection gain. This gain is based on a correlation coefficient between the two images. The results show its effectiveness in providing in an unique image both the geometric information and elastic properties of the investigated tissues.