When AI attacks multiple sclerosis

May 30 is World Multiple Sclerosis Day, an incurable disease that attacks the nervous system. Artificial intelligence may allow early diagnosis for better treatment skills aimed at slowing its progress.

An autoimmune disease, multiple sclerosis is characterized by degeneration of myelin, the membrane that protects the action of neurons. Communication between the nervous systems is then gradually disrupted, resulting in increasingly significant motor and nerve damage. Multiple sclerosis is currently curable, but treatment can relieve certain symptoms, especially if the disease is initially discovered. But today, multiple sclerosis is detected rather late.

Automatic wound detection with deep learning

Involved in the early diagnosis of multiple sclerosis, in particular, MRI observation of biomarkers, such as lesions or abnormal volumes of certain cerebral structures, Reda Abdellah-Kamraui, a doctoral student at the Bordeaux Computer Science Research Laboratory (LABRI), explained. It takes a long time to manually extract this information from MRI images, and so automated techniques have been created.. A

Image that combines the results obtained by different experts after two MRI observations conducted at intervals of a few months compared to those obtained by artificial intelligence.

Her thesis “Deep Learning for Big Data in Neuroimaging”, directed by Pierre Coupe, focuses on these questions. Methods of deep learning (or Acquiring deep knowledge), Designed for image detection work, naturally these complex and tedious activities are automatically used. ” Artificial Intelligence (AI) remains a tool capable of making mistakesReda Abdellah-Kamraui emphasized. Physicians retain the exclusive right to diagnose. Deeper learning, however, makes it possible to obtain an objective prediction, where two physicians do not necessarily give the same explanation.. A

Fake pictures for real AI training

In-depth learning involves the collection of examples and data, with which algorithms train themselves to distinguish important components of MRI images. But these components are not standard, because the manufacturers of different MRI devices and their models do not have the same rendering. In Bordeaux, Reda Abdellah-Kamraui is specifically studying the generalizations of neural networks so that they can train despite having different information.

An example of a lesion that was not identified by experts during the first phase of the test, but was identified by the segmentation module and accepted by all specialists during the second phase of the test. The first row shows the baseline FLAIR scan, the second FLAIR scan a year later. The red arrows indicate the wound of interest.

In the same vein, part of the work is dedicated to the generation of synthetic images, which compensates for the lack of data to train algorithms. Reda Abdellah-Kamraui took part in the Challenge of the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), dedicated to medical imaging, on new wound identification and segmentation due to multiple sclerosis. “ The same patient had to identify new lesions from two consecutive MRIsReda specifies Abdellah-Kamraui. The concern is that as soon as lesions are identified and patients are treated, subsequent MRIs will not show significant differences and so we lack the data to train our algorithms. We then propose a technique where we create fake MRI images that mimic the case of a patient who has not been treated for several years, then we use them for our AI training. A

Reda Abdellah-Kamrawi and her colleagues are again keen to predict scores of multiple sclerosis severity from MRI images, but also from demographic and clinical data. This score is a very important parameter for doctors.

Generalized education

For all of this work, the team uses Python computer language and employs a dedicated library that allows algorithms to read IRM. In addition to deep learning, researchers are developing transfer learning, which allows an algorithm to master a new task for which it was trained exactly. This may seem simple, but systems based on artificial neural networks often have to start from scratch, or almost, learn a new mission, although it may seem like the first one.

120,000 patients with multiple sclerosis in France

However, MRI studies go beyond the single framework of multiple sclerosis. Jose V from Polytechnic University of Valencia (Spain). Pierre Cope, the thesis director of Reda Abdellah-Kamrawi, made with Manjan volBrain. This platform makes it possible to download MRI data and perform many useful tasks automatically. Diagnosis of neurodegenerative pathologies, including multiple sclerosis, but also Alzheimer’s or Parkinson’s disease.

Boris Mansenkal, a research engineer at LABRI, has integrated these solutions. An ANR project, entitled Deepvolbrain, Also running the platform to adapt to the big data challenges, represented by the explosion of MRI data size. Doctors such as Thomas Turdias, a university professor at Bordeaux University Hospital and a hospital practitioner, are involved in the project...

Medical decision support

In France, other groups are working to help diagnose multiple sclerosis by AI. The MPEN team from the Institute for Research in Informatics and Random Systems (IRISA) is thus taking part in the Primus program, established by University Hospital Renaissance, which was inspired by eight million euros in calls for “Richter Hospital-Academic” projects. In health “.” Multiple sclerosis 120,000 patients in France, each of whom undergoes MRI every yearGilles Edan, a hospital practitioner and Professor Emeritus at University Hospital Rennes, is very involved with Primus. Add to that the fact that neurologists and radiologists cannot specialize in all the diseases under their discipline, so it is impossible for each MRI image to be interpreted by a specialist in multiple sclerosis.A

The Primus project focuses on two tools. The first will aim to support medical decision making, and to get patients to agree to the medication prescribed for them, as they may have significant side effects. This future tool will be based on very high quality data from a team of pharmaceutical company clinical trials as well as a team of closely monitored OFSEP patients. ” He finds patients who share the most features with us, then shows us how different treatments have worked on them.Gilles Edan continued. Then we have solid and personalized indicators of the possible evolution of the disease under treatment. A

Primus’ second tool focuses on MRI and, thanks to a large database, will provide references to doctors who do not specialize in multiple sclerosis. ” It is a real revolutionGilles Edan delighted.It’s like our memory and experience of treating 10,000 patients with just one click!

Read our site
Brain MRI has automatic translator

Leave a Comment