Harry Lin - Michele Guerreri - CMIC seminar series

Harry Lin - Michele Guerreri - CMIC seminar series

Harry Lin - Brief Introduction of Image Quality Transfer for Paediatric Epilepsy Diagnosis Assistance in Developing Countries: Advance, Technology, and Application
Abstract: Epilepsy has become the most common neurological disease for Nigerian children. In the developing countries including Nigeria, the open-bore 0.36 T MRI scanner is more popular due to easy installation and economic electric power consumption. However, this low-field scanner is not sufficient to examine the subtle legion. The legion visualisation in children’s brain may be hindered by low SNR and low contrast in between GM and WM.  The proposed image quality transfer (IQT) is a machine learning technique producing a high-field MRI image with the same image quality as if scanned by a high-field scanner. The IQT method incorporated with U-net, one of the convolutional neural network architectures, propagates the low-field image to the corresponding high-field one by learning a transferring map from input-output pairs in a training set. We propose several tricks to implement IQT suitable for 3D volumetric image on GPU computing realistically. The resulted images show a prospect of applying IQT to the clinical stage.

Michele Guerreri - Diffusion MRI models for brain tissues: a NODDI revision for multiple b-tensor encoding and γ-imaging to investigate brain-aging
Abstract: Diffusion MRI (dMRI) is an important tool to investigate the microscopic structure of living tissues non-invasively. The dMRI signal contains indirect information about the microscopic organization of the tissue under investigation. However, The reconstruction of the underlying microscopic architecture is not simple. A plethora of mathematical models have been proposed to address this problem.

In this talk I will present the part of my PhD work which focused on dMRI models for brain tissues. The presentation is divided in two sections: in the first part, the revision of NODDI model will be discussed. NODDI is a widespread biophysical model which aims to infer neurite morphology from conventional dMRI data. NODDI has been shown to be incompatible with data acquired using a new generation of dMRI acquisitions, known as multiple b-tensor encoding. NODDI’s tortuosity constraint has been addressed as the cause of this inconsistency. The aim of this work is to revise the NODDI model in order to make it compatible with multiple b-tensor encoded data, showing that this can be achieved without forgoing the tortuosity constraint.

In the second part of the presentation, the assessment of age-related brain modifications via the so called γ-imaging approach will be discussed. γ-parameters are thought to depended on local inhomogeneities due to magnetic susceptibility differences between tissues and diffusion compartments, potentially providing information about myelin orientation and iron content within brain. In this study, we analysed a cohort of young-to-elder healthy subjects. The results showed that γ-imaging provides complementary information about brain aging microscopic modifications compared to other conventional dMRI techniques.