Efficient, modular tools to aid interpretation of medical images
© École Centrale Paris
Medical imaging, scans and MRIs provide a lot of data, but identifying the development of an illness or detecting a slow-growing tumour is no easy task. The Galen team is developing algorithms designed to help doctors identify these changes as early as possible. The team's work focuses essentially on methodology and has been successfully applied in a number of contexts, as demonstrated by the significant contribution of its researchers to MICCAI’2012 and previous symposia.
Nikos Paragios, Galen team leader, Inria Saclay–Île-de-France research centre
You are presenting four projects at MICCAI 2012: what do they have in common?
Our contribution to all four articles is a methodological one. We design intelligent programs capable of helping doctors to interpret medical data better in order to identify and treat illnesses. Over the last ten years or so, we have developed a general data analysis framework based on graph theory which can be used to produce solutions offering a good level of optimality and short calculation times. It can also be applied to problems concerning different types of data, illnesses or populations. The work has a direct impact on practices, as we are solving scientific problems that are very close to clinical preoccupations. Our solutions have given rise to technology transfers and are already being used in hospitals.
And two of your contributions directly concern the interpretation of patient images?
The first project, conducted in conjunction with Montpellier University Hospital and the company INTRASENSE, concerns the detection of brain tumours and, more specifically, type II gliomas, using data provided by patient scans or MRIs. Our approach makes it possible to establish correspondences between a diseased brain and a healthy brain so as to detect any anomalies, and to establish the timing of these correspondences in order to monitor the changes in the tumour. The tools that we have developed help the doctor to identify the tumour and to decide on the right time to operate.
A second project, conducted with doctors from Henri Mondor and La Pitié Salpêtrière Hospitals as well as the Institut de Myologie, applies the graph methodology to measure the muscle volume of patients suffering from myopathies. These diseases cause muscle to decay and be replaced by fat. By establishing correspondences in time between scans or MRIs, doctors can not only track the development of the disease but also identify the affected muscles in order to better target treatments, such as gene therapy.
What are the intended applications of the other two projects you are presenting at MICCAI?
One of them applies the methodology to cell populations in an experiment conducted by biologists and computer scientists at Houston University. In this case, graph theory is used to identify the active genes in a mouse cell by establishing correspondences with a base model.
In the last project, the methodology plays a less incongruous part. General Electric must develop ultra-fast scanners adopting a smart sensor motion system that reduces acquisition time - and thus the time for which patients are exposed to radiation - but still produces the same quality of image. To this end, sophisticated reconstruction tools have been developed (sparsity/parsimonious methods) which will use graph theory to take into account the movements of the patient or of organs such as the heart, in order to reduce reconstruction artefacts.