Medical image analysis: a subject for the future?
© Christof Seiler
We spoke to Christof Seiler, a doctoral student in the ASCLEPIOS project-team in Inria Sophia Antipolis (near Nice) and the ISTB of the University of Bern in Switzerland, whose advisors are Xavier Pennec (Inria) and Mauricio Reyes (ISTB). Christof’s research focuses on medical image analysis, and in particular analysing 3D images of human mandibles. He is presenting a paper at MICCAI 2012, for the second year running, this time on “Simultaneous Multiscale Polyaffine Registration by Incorporating Deformation Statistics”.
Medical imaging has been around for over a hundred years now, but modern 3D techniques, such as CT scans and MRI, are a far cry from the 2D X-ray methods that were the mainstay for most of the 20th century. CT scans and MRI yield much more information than simple 2D scans and, as a consequence, there are vast amounts of data to analyse. The scans can also be taken over time to analyse how a patient’s situation evolves, something that introduces a temporal aspect to the images.
Helping medical practitioners
Our goal is to help medical practitioners analyse these 3D images,” explains Christof. “Instead of simply looking at the images and inferring certain details, these images can now be analysed automatically thanks to computer programmes and algorithms based on sophisticated mathematical theories.
Christof’s work involves analysing images of mandibles in an effort to ultimately design better implants for patients that require reconstructive surgery. The human mandible is a complex structure that varies greatly on many length scales – from centimetres to microns. Christof and his colleagues therefore employ so-called locally affine (polyaffine) image registration methods that can pick up non-linear deformations across this range of length scales. This approach is quite different to traditional analysis techniques that rely on sequential coarse to fine registration to decipher multiscale deformations.
The researchers analyse thousands of 2D image slices taken from around 40 patients. The models are based on statistical analysis of the images and involve a set of parameters whose correct values need to be defined. “A lot of our work depends on how powerful the computer we are using is ,” explains Christof. “Analysing the ‘answers’ produced by the computer are a challenge in itself because we need to make our results intelligible to medical practitioners, who usually have much less experience in computers or programming .”
Christof says he is lucky to be studying at two different institutions at once, in Sophia Antipolis and Bern. The work is interdisciplinary and involves computer science, geometry, topology, statistics and probability theory. “It is really interesting to see how much mathematics is actually involved,” he says, “and everyday, I have the impression of tackling and learning about a different subject, which makes my research extremely diverse .
I very much wanted to work at Inria because of this interdisciplinary aspect, and the fact that collaborations between the different disciplines are so strong here. My supervisors, colleagues and myself constantly liaise with each other and share ideas that we can all follow up afterwards, independently or as a team. It’s often hard to say who came up with such and such an idea, because we work so closely together!”
Ces articles peuvent vous intéresser :
- Médecine numérique MICCAI invente la médecine de demain
- © Inria - Robert Palomba Médecine numérique Personnaliser les modèles du cœur pour mieux cibler les thérapies
- Longitudinal atrophy flow in Alzheimer's disease Médecine numérique Neuroimaging and computing for better diagnosis and treatment of Alzheimer’s disease
- Médecine numérique Des outils pour mieux tirer parti de l’imagerie cérébrale
- © École Centrale Paris Médecine numérique Des outils modulaires et efficaces d’aide à l’interprétation des images médicales
- © Tonis Pan - Fotolia.com 21/11/2012 Rencontre Inria Industrie - Santé
MICCAI sur Twitter
Suivez la conférence sur Twitter.