À la Une

Soutenance de thèse Matthias Becker


M. Matthias Becker soutiendra, en vue de l'obtention du grade de docteur ès sciences, mention informatique, sa thèse intitulée

Efficient Extraction of Musculoskeletal Structures from Multi-Channel MR Images

  • Sous la direction de: Prof. José Rolim
  • Et la codirection de: Prof. Nadia Magnenat-Thalmann


In a society with a growing share of elderly, musculoskeletal diseases (MSDs) have become a major burden not only for those affected but also for the health care system and have been found to be the main reason for work time loss in the European Union. MSDs are a group of disorders, including osteoarthritis, which affects body parts like bones, muscles, tendons and joints. They cause pain and reduce mobility and flexibility. A better understanding of these diseases can be achieved by acquiring knowledge of the patient-specific morphology of the involved anatomical structures. Magnetic resonance imaging (MRI) is a non-invasive technique for acquiring volumetric tomographic images. MRI, like other medical imaging techniques, is subject to constant improvement. This concerns, among others, image quality, resolution, acquisition duration and costs. It has led to an ever-growing number of acquisition with higher resolutions, resulting in big numbers of large data sets. The interpretation of MRI data is a lengthy process that requires special training and knowledge about anatomy and imaging properties. Depending on the acquisition parameters, analysing the MR images, e.g. to identify organs or pathological tissue, is a complex and time-consuming task. With higher resolutions, an increase in the total number of scans, and a reduction of medical personnel, there is less and less time left for image interpretation. This requires automated processing of the data to assist the medical personnel. Some MRI protocols can generate multi-channel images, which can highlight different anatomical structures in different channels.

We focus on the extraction of musculoskeletal structures (bones, muscles) from the lower limb. In our work, we have focused on three main objectives: development of an MRI protocol, processing and labelling of the image data, and the exploitation of multi-channel data during the segmentation of individual muscles. All these tasks have been addressed with a focus on efficiency. We propose an MR acquisition protocol that generates seamless, high-resolution images of thigh and calf. The acquired image data is then processed, following our pipeline. This includes stitching, noise and bias field reduction and the identification of different classes of image content. We present a method to identify air and muscle tissue in the image and massively parallel version of the approach.

To identify the individual muscles, we align a muscle template. This template is modified with our deformable model framework under the influence of image forces to match the actual anatomy. This is an iterative process and the image forces use the multi-channel image data to find relevant image features. We propose coupled multi-resolution deformable models that allow working on different resolutions in parallel. This work is aimed at helping the understanding of diseases by providing personalised anatomical models. There is a large number of applications; we show a contribution towards the digital patient and the determination of personalised pennation angles.


Date: Mardi 20 septembre 2016 à 10h30

Lieu: Battelle bâtiment A - Salle de cours 316-318 (2ème étage)

7 septembre 2016
  À la Une