AAN 2026 | Using immune signatures and machine learning to improve neurological diagnosis
Raphael Bernard-Valnet, MD, PhD, University of Lausanne, Vaud, Switzerland, discusses research investigating disease-specific inflammatory patterns to improve neurological diagnosis. He explains how cerebrospinal fluid proteomic and inflammatory markers, combined with machine learning approaches, may help distinguish clinically overlapping neurological diseases and identify atypical patient profiles. This interview took place at the 78th American Academy of Neurology (AAN) Annual Meeting in Chicago, IL.
These works are owned by Magdalen Medical Publishing (MMP) and are protected by copyright laws and treaties around the world. All rights are reserved.
Transcript
We basically try to see if adding some proteomic markers to what we usually do for lumbar puncture, so the markers we usually measure in lumbar puncture, if adding some proteomic markers that are related with inflammation mainly, so cytokines, chemokines, and growth factors is able to segregate neurological diagnoses behind against each other so to do that we first looked at about 60 markers in the CSF of 800 patients a bit less have been kept in the final analysis, around 750...
We basically try to see if adding some proteomic markers to what we usually do for lumbar puncture, so the markers we usually measure in lumbar puncture, if adding some proteomic markers that are related with inflammation mainly, so cytokines, chemokines, and growth factors is able to segregate neurological diagnoses behind against each other so to do that we first looked at about 60 markers in the CSF of 800 patients a bit less have been kept in the final analysis, around 750. And among these patients, ranging from multiple sclerosis to Alzheimer’s disease, encompassing many other diagnostics, like migraine patients, it’s basically all the spectrum of patients that have received a lumbar puncture, we are able to show that we have immune signatures that are different from one group to another, yet it’s still a bit difficult to predict the diagnosis itself in a huge organized manner. If we use, for instance, machine learning tools, but yet we’re still able to differ between diagnoses that are close, that look like clinically the same because in clinics we do not want to separate I don’t know Alzheimer’s disease from multiple sclerosis it’s usually not a question we have but if we want to separate for instance multiple sclerosis from neurosarcoidosis that can present like here we have the impression that this panel can be of, and we have shown that it adds something to what is existing with the usual markers, and it can go towards a bit more personalized medicine, and it can help the diagnosis. So we’re still working on that with many options to offer visualization to the clinician and also to provide edited reports. And the last thing we’d like to work also on is to try to define if the patient is an outlier, knowing, for instance, I know this is a relapsing-remitting patient, at least it’s my final diagnosis, but still I am unsure. Is the machine learning system or is the prediction of this multi-omic measurements telling us is it multiple sclerosis or not. And we would have a warning and a red flag for some diagnoses to say okay to the clinician we should go a bit further because it seems that your patient is an outlier.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
Read more...
