Read the article." /> Read the article." /> Read the article." />
Skip to Content


MDS makes every effort to publish accurate information on the website. "Google Translate" is provided as a free tool for visitors to read content in one's native language. Translations are not guaranteed to be 100% accurate. Neither MDS nor its employees assume liability for erroneous translations of website content.

International Parkinson and Movement Disorder Society
Main Content

Using EEG to Identify Parkinson's Subtypes

December 04, 2023
Dr. Sara Schaefer discusses with Prof. Marc Vérin and Prof. Mahmoud Hassan their recent paper in Movement Disorders Journal on the use of Resting-State High-Density EEG to neurophysiologically stratify Parkinson's patients into subtypes in an effort to predict cognitive and motor progression, properly evaluate for invasive therapies, and inform future clinical research. Read the article

[00:00:00] Dr. Sara Schaefer: 

Welcome to the MDS podcast, the official podcast of the International Parkinson and Movement Disorder Society. I'm your host, Sara Schaefer from the Yale School of Medicine. And today we'll be speaking with Marc Verin, a professor of neurology at the University of Rennes in France, and Mahmoud Hassan, CEO and founder of MINDIG in Rennes as well, an adjunct professor of the School of Science and Engineering at Reykjavik Iceland University. We're going to be talking about a paper that they published in the August 2023 issue of the Movement Disorders Journal called Identification of Parkinson's Disease Subtypes from Resting State Encephalography. Thanks for joining us.

View complete transcript

[00:00:47] Prof. Marc Verin: Thank you, Sara.

[00:00:47] Prof. Mahmoud Hassan: Thank you for having us.

[00:00:49] Dr. Sara Schaefer: All right, let's start with the background. What research is already out there on EEG in Parkinson's disease and what gap were you hoping to fill with this research? What are the [00:01:00] clinical and research related motivations for this work? Marc?

[00:01:04] Prof. Marc Verin: I'm specialist of movement disorders, and I'm using deep brain stimulation since about 30 years. And our problem is in clinical point of view to choose the best patient for DBS and in fact for STN-DBS subthalamic nucleus stimulation. We choose, patient without any cognitive decline.

That is to say if possible, pure dopaminergic disease. And the problem is that when we look at cognitive performance of the patient, some patient have very good performance, normal performance. And when we perform the deep brain simulation, we observe a few months after surgery that there is a severe sometime cognitive decline. The problem is in our point of view is the neuroplasticity, because in this case, the patient seems to be normal, seems [00:02:00] to be pure dopaminergic. But there is already all diffuse disease. And the problem now is to see what the clinic is not able to see. That is to say, to see the neuroplasticity and the readjustment of the networks in the brain. In order for the patient with diffuse disease to maintain a normal quantitative performance. And that is the reason we developed with my colleague Mahmoud a new tools to see the functional network in order to see the modification of the network, even the patient have a no cognitive decline.

[00:02:39] Dr. Sara Schaefer: So you mentioned in the paper the importance of using this type of information for prognostication, and now you're saying also for prognostication in terms of trying to figure out if a patient would benefit from DBS or if DBS might actually worsen their cognitive situation. And then also the other thing that you mentioned was, [00:03:00] obviously, we're always trying to stratify people well for clinical research, right?

[00:03:05] Prof. Marc Verin: If you look at the new neuroprotective treatment, of course it's mandatory to select the best patient the best treatment. And if you include patient with inapparent , pure dopaminergic disease and In fact, if a person has a diffuse disease it's a bias, of course, for the study of neuroprotection.

And the study may be false without good results, despite, in fact, a very good effect. Because we have in the same groups very dopaminergic cured patients and patient with diffuse disease. And that is the reason for us, HD-EEG is also able to select and stratify the patient in order to develop new tools of neuroprotection.

Not only for DBS, in the near future, for the neuroprotection treatment.[00:04:00] 

[00:04:00] Dr. Sara Schaefer: So what research is already out there on EEG and Parkinson's disease? And how did you use that to design your study? Mahmoud?

[00:04:09] Prof. Mahmoud Hassan: Yeah. Sure. So EEG is electroencephalography. The idea of EEG is to record the electrical activity of the brain using some electrodes that we put on the scalp, so it's noninvasive. The main advantage also that it can be portable. And also one of the main advantage of EEG is the excellent time resolution.

So we can record activity at the millisecond. So EEG, it exists like since 100 year now, it was like 1924. And it was, used mainly in epilepsy. Because on the EEG recordings, we can see clearly the seizure. We can see the epileptic spike, the epileptic activity, even visually without any processing of the EEG.

So EEG is a very complex signal, but in the case of epilepsy, it's easy to see it. For this reason, so far, I mean, EEG is mainly used in the clinical application for epilepsy. But recently, [00:05:00] we started with the progress of the technology, with the progress of EEG, with the progress of the EEG analysis, we started to see EEG used in many, many different brain disorders, including neurodegenerative diseases, but also psychiatric disorders.

The question is, if I show you a EEG of Parkinson's disease patient and healthy control age match there is no difference. I mean, visually, there is no difference.

So the question is how to extract knowledge, how to extract biomarker from this EEG that can differentiate between both. And here is the data processing part. So far several studies showed that there is a difference in term of EEG activity between Parkinson's disease patient and healthy controls, mainly in the frequency domain, which mean we have like more what we call low frequencies in patient compared to healthy controls. So we have this shift into low frequencies. So if you adjust to describe EEG we have what we call the rhythms. So it's called the rhythm. We have different like in our [00:06:00] waves in the EEG called theta, alpha, beta and gamma. in the case of Parkinson's disease patient or in general neurodegenerative diseases, we have a slowing of that EEG activity.

So we have more low frequencies compared to healthy controls. So what has been done so far is like what we call the group analysis. I mean, comparing two groups, healthy controls compared to a patient. Interesting in terms of physio pathology, understanding the difference, but it's not enough because it doesn't take into account the heterogeneity that can exist within the patient population. So what we try to do is to take into account this heterogeneity and go from group average analysis into subgroups using this, method to extract subphenotypes. So this is a main, difference in contribution of the paper compared to what already exists.

[00:06:52] Dr. Sara Schaefer: And can you explain for the uninitiated the difference between an EEG that you might order for somebody who you suspect has a seizure disorder [00:07:00] versus the high density EEG that you used in your study?

[00:07:03] Prof. Mahmoud Hassan: EEG that it's used in seizure can be like a standard EEG, which is like 1932 channels on that because we can see, the EEG clearly I mean, that, the activity, the seizure activity one sense, like maybe 10 more or less 10 years now, we had what we call high density EEG.

And I mean, the idea of high density is to put more electrodes on the brain. So more electrodes on the scalp, it means we are capturing more information. So we have higher spatial resolution. This high density EEG Is crucial for advanced data analysis because with this high density EEG, we can go deeply into the cortex.

We can from technical term is what we call source localization. So we go from scalp. into the brain. We reconstruct the brain sources. So we need high density EEG to go deeply into the cortex. So to reconstruct the brain sources. So this is the main advantage of [00:08:00] having high density EEG compared to standard EEG.

[00:08:03] Dr. Sara Schaefer: Three patients in the study were excluded due to artifacts. I was wondering as I was reading the paper, was this related to PD symptoms like tremor or dyskinesias? Are there certain types of patients who have those extraneous movements for whom this test may not be as helpful?

[00:08:22] Prof. Mahmoud Hassan: Yeah, one of the steps of when we analyze the EEG. We called it pre processing, which means that we have to remove what we called artifacts and noise from the signal before going further into the analysis. So one of the artifact is eye blink. So people who blink a lot.

It's an artifact that should be removed. So we have some tools to do it. One of the artifact is electrical environment. So we have like a lot of machines and stuff. It's artifact muscle artifact. People who try to do something like that. So there is some muscle artifact. So it's not easy.

And we have a lot of movement during recording. [00:09:00] And based on our tools, we have some criteria to select. It's really noisy, a lot of noise, more than a signal. And then we remove this patient. So this patient were removed because of the noisy data, because we cannot process noisy data.

[00:09:15] Dr. Sara Schaefer: Were those people eliminated because of PD related symptoms, or not?

[00:09:21] Prof. Mahmoud Hassan: No, it's more related to , the signal quality.

[00:09:24] Dr. Sara Schaefer: All right. Your analysis looked at five different frequencies from delta up to beta in four brain networks and used this information to identify three Parkinson's subgroups. Can you explain what you found?

[00:09:39] Prof. Mahmoud Hassan: Yeah, this analysis, it's called unsupervised cluster. Unsupervised, it means that we didn't give the machine or the algorithm any a priori about the cluster. The cluster is the subphenotypes. Clustering because we found some subgroups in the patients.

The idea is that we have this machine, I mean this clustering machine, this intelligent machine. [00:10:00] We give this machine all the data that we have. So in our case, we give them data for EEG, from the spectrum, spectrum frequency analysis, , the, the rhythm of the EEG waves. And this machine is able to find some clusters within this patient. And using this different EEG waves and networks we found or the machine told us that there is a three clusters in our data and these, clusters have different signature. Different. fingerprint in terms of EEG frequency. So there is like this clearly slowing. There is one group with really slowing waves. They have very, very high, low frequencies, like theta band, for example, and delta band. And some of them, they have different activation of specific brain networks. This is also one of our really know how in the team with Marc is how to really identify resting state networks and develop what we call network based [00:11:00] biomarkers for Parkinson's disease.

So we have something like mixed in terms of spectral, but also network fingerprint of each sub cluster. So we found three sub clusters. Three sub phenotypes with different spectral and connectivity behavior.

[00:11:16] Dr. Sara Schaefer: So you mentioned the more diffuse, more cognitively impaired group. That's the G3 group, right? And you talk a lot about the differences between G1 slash 2 and G3 where G1 slash 2 is more of the motor predominant subtype. Can you discuss a little bit more about the differences between G1 and G2 the two motor predominant, non diffuse subtypes, and why did you decide to make this distinction?

[00:11:45] Prof. Marc Verin: In clinical point of view we observe first the study we are able to analyze the data at baseline after three years and after five years. And the important point is that at baseline, [00:12:00] none of the patient have cognitive decline, but after five years, some patient have no cognitive decline. Some patients have mild cognitive decline and some patients have severe cognitive decline. And the clinical point of view is overlap with electrophysiological data and we can in fact distinguish at baseline. We are able to predict at baseline Even if there is no cognitive decline, this patient has a future diffuse disease. This patient have a future intermediate disease, mild connective decline, and this patient have a future pure dopaminergic disease, poor motor disease, in fact with no cognitive decline. Even if at the beginning, at baseline, there is no difference in terms of cognitive decline.

But that is the reason for with this tool we are able to see at baseline, what the clinic is not able to see and to predict the future. For us, it's a very important point [00:13:00] in order to, at T0, to choose the best treatment for the patient. If there is the proof, that it is a future diffuse disease or intermediate disease, we do not choose DBS and we prefer apomorphine pump or levodopa intestinal gel, for example.

At this time, it's very important for us to decide the best treatment for the future. And also, of course, in the near future for the neuroprotection studies.

[00:13:31] Dr. Sara Schaefer:  Those of us who practice in the inpatient setting certainly have come to know that slower frequencies in the delta and theta ranges, mean encephalopathy. We see that all the time, right? And you found that those with more cognitive impairment in the G3 cohort generally showed more power in those slower frequencies.

What about the faster frequencies? What do you make of the high beta power in the somatomotor network in the G1 cohort, for example?

[00:13:58] Prof. Mahmoud Hassan: Yeah, I think this is a [00:14:00] very, very good question. I think one of the decisions that we made is to don't take any a priori about the frequency band or the network involved in the analysis. Sometime when we have, , hypothesis driven analysis, we can focus on one frequency band, which is good.

But sometimes we miss a lot of other information and other frequency bands. So we, we know for a long time now this slowing of the frequency in neuro degeneration. But we don't know a lot in terms of a brain network or functional brain network, what's going on for the season in our case, we used all the information that we have without any a priori.

So we use all the frequency bands. Say from theta alpha beta, but also we used all the resting state networks, including default network, frontotemporal networks and so on. I mean, what we got let us think that the problem is much more complex than what we thought I mean, it's not a frequency band.

It's not one network, but it's a mixture of very complex pattern that [00:15:00] characterize each group. So, I mean, we were really happy with this sub clustering, but still a lot of things that we don't understand yet. Because why we have like a high activation in term of... default mode network less than frontal temporal network and some specific frequency band for this group. Not for the other one. This still not really easy to understand because the pathology is not easy to understand. And we think that it's even much more complex than that. So I think one of the message of the paper is to embrace this complexity in term of spectral, in term of network.

And for the reason we see that the patterns are really complex. It's not really related to one specific frequency band or one functional network.

[00:15:41] Dr. Sara Schaefer: Marc, it seems that we're moving away from the more classic distinction, phenotypic distinction of Parkinson's as tremor predominant versus aconetic rigid, right? That's how I learned it way back when towards the diffuse versus motor only subtypes and your [00:16:00] stratification seemed to go along these lines.

Can you talk about this a little bit more? Did your data support any value in tremor predominant versus aconetic rigid classifications?

[00:16:12] Prof. Marc Verin: In fact, there is no effect of the motor phenotype in this group on the cognitive evolution and in the first paper in disorders in 2022, we have studied all the patient, all the database with 77 patients and we observe no correlation between the motor phenotype and, the functional networks.

[00:16:38] Dr. Sara Schaefer: Very interesting. So getting to practicality here, how did your team actually learn to analyze this data? Are there epileptologists on the team? For those movement disorders folks who are listening, who are not epileptologists, how might one engage with this line of research?

[00:16:55] Prof. Mahmoud Hassan: I think well, epileptologists are very interesting to read [00:17:00] the EEG even visually , to read the EEG and to have some visual inspection on it and to understand really the EEG in our case, we need the people who know how to do data science. How to really extract features from the EEG. The team is mainly people who know how to extract knowledge from EEG.

So it's more our team now at MINDIG and the team that we have with Marc. So Marc is more the clinical referee for Parkinson's and how we can understand and interpret and the people who analyze the EEG. So, I mean, for me I work with EEG for more than 10 years now.

And the main know how that we have is how to analyze this EEG. It's not easy at all. It's much easier in the case of epilepsy because we see things on the screen, on the signal. In the case of Parkinson's, Alzheimer's, but also in psychiatry, the EEG is really, really increased. I mean, there is an increased use of EEG in psychiatry because the idea is to really, to develop, biomarkers, I mean, from [00:18:00] EEG.

It can be for cognitive decline in our case, it can be for cross disorder in the case of psychiatry, for prediction, for diagnosis, for follow up. But the question is how to extract the right patterns from EEG for the right questions. So this is really the hardest part.

And this is a know how that we have. And the main skills to do this is signal processing and data science. 

[00:18:25] Prof. Marc Verin: I'm not a epileptologist . And for me HD-EEG is now a new neuro imaging tool, in fact, like fMRI, but with a better temporal resolution, of course, in comparison with fMRI. fMRI is a about the best. It's one second. What's happened in the brain in one second?

With HDEEG we're able to see with one milliseconds and the most important point. Also, we have published with the resting sets we met, but we're able to study the dynamic of the brain, and [00:19:00] probably it is more sensitive to perform cognitive tasks during recording of EEG. And during recording of EEG, we are able to see the networks in action.

And probably it is more sensible to detect neuroplasticity even if with the rest in text, you are already able to differentiate the type of separation. We probably with the dynamic of the brain, we will more able to detect very early in the evolution.

[00:19:30] Dr. Sara Schaefer: And also quite a bit more accessible than something like fMRI. From an international perspective and a socio economic perspective. 

[00:19:38] Prof. Mahmoud Hassan: From economic perspective is much, much cheaper. And also can be and will be, I mean, tomorrow, I mean, we'll be portable and then we can do at home recordings. I mean, this is, this is a near future, not really the long future for the EEG. So it's, much cheaper. It's a direct measure of the neural activity.

Also, this is very important. And also it's non invasive and with [00:20:00] what Marc mentioned and it's excellent time resolution. One of the factor that the use of EEG in different disorders other than the epilepsy is because sometime that people are afraid of the EEG, which is normal because in term of data collection and mainly in term of data analysis, it's not easy at all.

Well, this is what we try to do at MINDIG because at MINDIG our idea is really to help all the neurologists and even psychiatrists interested in the EEG. That we are the people who analyze EEG, who tried to extract biomarkers and really I mean, asking the right question by neurologist or psychiatrist and the data analysis part can be done by expert.

I think it's changed a lot compared to a few years ago, but this is one of the reason why EEG was not that popular so far.

[00:20:47] Prof. Marc Verin: am a neurologist, but... I have also a PhD in neuropsychology, in fact, and when I've seen the results of the EEG I think we have to revisit all the confusion made with fMRI, [00:21:00] in fact, because now we have dynamic tools. We have to revisit, I think, all the neuropsychology with these new tools.

[00:21:10] Prof. Mahmoud Hassan: I think just to add to this, I think it's very complimentary. I think that the fMRI has excellent special resolution. EEG has an excellent time resolution. I think that people who try to do some multi modality, , experiment is really a good way to combine both techniques, for one question.

[00:21:29] Dr. Sara Schaefer: Okay. So people don't need to be in one camp or the other. We know what camp you guys are in. All right. So we talked about stratifying patients with Parkinson's disease in terms of how they might progress moving forward even from the early stages. Do you anticipate that we may be able to differentiate Parkinson's disease from atypical parkinsonisms, which may be hard to differentiate at very early stages in the future. Marc?[00:22:00] 

[00:22:00] Prof. Marc Verin: Yes, it's a very good question. We have a study with our colleague from Antifa, Martinique and Guadeloupe in order to differentiate atypical parkinsonism and Parkinson's disease, in fact. And also not only in the French Indies, but also PSP and MSA. And of course, because of there is a different neuronal loss in this disease. Probably there is a very clear difference in terms of networks and modifications of the networks.

It is future goal. 

[00:22:33] Prof. Mahmoud Hassan: Yeah, and we can ask also, Marc, the reviewers of our grant to be kind with this proposal. We are waiting for this project. So, yeah, we can ask them to be kind.

[00:22:44] Dr. Sara Schaefer: I promise to our listeners that they did not plant this question.

[00:22:48] Prof. Marc Verin: Yeah.

[00:22:50] Dr. Sara Schaefer: All right. Thank you both. Anything else to add before we close out?

[00:22:54] Prof. Marc Verin: Thank you. For you to be interest for you with HD-EEG. For me it's a revolution because I [00:23:00] am movement disorder specialist, and I'm not epileptologist, but it's difficult to convince now or colleagues in parkinsonologist to be interested in EEG, but I think the two papers in movement disorder may modify the view of EEG for the specialist of Parkinson's disease.

I hope so.

[00:23:21] Dr. Sara Schaefer: Well, you've got me very interested. Mahmoud?

[00:23:24] Prof. Mahmoud Hassan: Well, I mean, thank you so much for the invitation. And I think it's interesting for us. I mean, to share our thought about these papers. I think it's a long way. And the idea is high density EEG is one of the tools. I mean, in addition to the other tools that can give another window to the brain activity and also maybe answer to some question that not possible to answer with other techniques.

We think that high density EEG has combined with the right tools, analysis. Now we went for this clustering. Now we are developing some tools that combine EEG with machine learning, for example, for classifying, but [00:24:00] also predicting , some specific aspect in Parkinson's disease and other diseases.

So we believe that, yeah, high density EEG, when it's analyzed well and right, has a really high potential in different brain disorders, including PD.

[00:24:13] Dr. Sara Schaefer: All right, thank you very much.

[00:24:15] Prof. Mahmoud Hassan: Thank you, Sara. 

Special thank you to:

Prof. Marc Verin
Professor of neurology
University of Rennes
Rennes, France

Prof. Mahmoud Hassan
CEO and founder of MINDIG, Adjunct professor
School of Science and Engineering, Reykjavik Iceland University



Sara Schaefer, MD 

Yale School of Medicine

New Haven, CT, USA

We use cookies to give you the best possible experience with our website. These cookies are also used to ensure we show you content that is relevant to you. If you continue without changing your settings, you are agreeing to our use of cookies to improve your user experience. You can click the cookie settings link on our website to change your cookie settings at any time. Note: The MDS site uses related multiple domains, including and This cookie policy only covers the primary and domain. Please refer to the MDS Privacy Policy for information on how to configure cookies for all other domains on the MDS site.
Cookie PolicyPrivacy Notice