Multiple sclerosis (MS) is an autoimmune disease of the central nervous system (brain and spinal cord), World Day 2022 dedicated to this incurable disease took place last May 30th. To diagnose it, doctors use a painless magnetic resonance imaging (MRI) examination of the brain and / or spinal cord. The deep learning could allow doctors to make an earlier diagnosis and thus put in place a more effective treatment to slow the progression of the disease.
An autoimmune disease, multiple sclerosis is characterized by the breakdown of myelin, the membrane that protects the axons of neurons. Specifically, the immune system, involved in the fight against bacteria and viruses, attacks myelin, the protective sheath of nerve fibers, disrupting communication within the nervous system, causing more and more motor and neurological damage. . 120,000 people are affected by MS in France, including 700 children, 3/4 of whom are women; 3,000 new cases are diagnosed each year, most often between the ages of 25 and 35.
If MS cannot be cured, new treatments can slow the progression of the disease today, and it will be all the more effective if the diagnosis is made early, which is rarely the case.
Automate the detection of lesions with the deep learning
The thesis « Deep learning for big data in neuroimaging », led by Pierrick Coupé, CNRS research director at LaBRI (Bordeaux Computer Research Laboratory), led by PhD student Reda Abdellah-Kamraoui since 2019, is at the heart of these questions. The methods of deep learningdeveloped for image recognition tasks, have been used to automate these complex, time-consuming operations to develop a new generation of quantitative MRI analysis methods that can cope with the rise of BigData in neuroimaging.
Reda Abdellah-Kamraoui explains:
“Early diagnosis of multiple sclerosis includes biomarkers, such as lesions or abnormal volume of certain brain structures.. Manually extracting this information from MRI images takes a considerable amount of time, so automated techniques have been developed.. »
« Artificial intelligence (AI) remains a misleading tool. Doctors retain a monopoly on diagnosis. Deep learning, however, provides an objective prediction, where two clinicians do not necessarily give the same interpretation. »
Generate false images to drive algorithms
The deep learning requires a wealth of examples and data to, in this case, lead algorithms to distinguish important elements on MRI images. The problem with Reda Abdellah-Kamraoui came from the fact that these elements are not standardized because the different MRI machines do not have the same renderings depending on their manufacturer and model. He therefore applied the generalization of neural networks, which makes it possible to drive algorithms despite heterogeneous data.
With the same idea, some of his work was devoted to the generation of synthetic images, so he was able to address the lack of data to drive the algorithms. In this context, Reda Abdellah-Kamraoui participated in the challenge of the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), dedicated to medical imaging, on the detection and segmentation of new lesions due to sclerosis in plates.
“New lesions had to be detected from two successive MRIs of the same patient. The concern is that because patients are treated as soon as lesions are spotted, the following MRIs will not show significant differences, so we lack data to drive our algorithms. We then proposed a technique where we generate false MRI images that simulate the case of a patient who has not been treated for several years, and then we use it to train our AI. ”
Together with his colleagues, he is also interested in predicting the severity of MS, from MRI images, but also demographic and clinical data. This score is a very important parameter for the adaptation of treatment by doctors.
Generalizable MRI automation
For all this work, the team mainly uses the Python language and a dedicated library that allows algorithms to read MRIs. In addition to the deep learning, researchers are developing transfer-based learning that allows an algorithm to master a new task through skills learned from previous tasks. Indeed, systems based on artificial neural networks often have to start from scratch, or almost, to learn a new mission, even if it seems similar to the first.
In addition, this thesis work integrates the tools developed within the volBrain platform.
The volBrain platform
The study of MRI is not just about multiple sclerosis. Pierrick Coupé, Reda Abdellah-Kamraoui’s thesis supervisor, co-created with Jose V. Manjón of the Polytechnic University of Valencia (Spain) the volBrain platform with 3,000 users worldwide and has already processed 140,000 MRIs. This open access platform allows you to download MRI data and automatically perform many useful tasks for the diagnosis of neurodegenerative diseases such as MS but also for Alzheimer’s or Parkinson’s disease.
The integration of these solutions is carried out by Boris Mansencal, a research engineer at LaBRI. An ANR project called DeepvolBrain is currently underway to adapt the platform to the challenges of big data due to the explosion in MRI data size. Doctors like Thomas Tourdias, a university professor and hospital practitioner at Bordeaux University Hospital, are involved in the project.