AI and EEG

[DerpyNeuroMD]

I'm finishing up a manuscript and I'm finding lots of recent research (i.e. last 1-2 years) showing machine learning can solidly interpret EEGs as well as, or better than humans. In my experience with waveform analysis and machine learning it seems to point towards the possibility of reliable automation within a few years. Plus we're seeing multifaceted models that include actigraphy, AI interpretation of video monitoring, ECG, and other features. It looks to me like deep learning will have this on lock very soon. As someone interested in epilepsy/neurophys, it has me wondering what the future holds. Granted this tech will need to be vetted. But it only took neuropace about 5 years. I am seeing a lot of other upcoming automation in other neuro fields, including stroke. I haven't heard a lot of talk about this and I'm wondering your opinions.

---

Members don't see this ad.

 

[Telamir]

I just went outside and confirmed that the sky…is indeed…falling.

That’s all for today’s forecast. Stay classy, San Diego.

 

[xenotype]

I'm finishing up a manuscript and I'm finding lots of recent research (i.e. last 1-2 years) showing machine learning can solidly interpret EEGs as well as, or better than humans. In my experience with waveform analysis and machine learning it seems to point towards the possibility of reliable automation within a few years. Plus we're seeing multifaceted models that include actigraphy, AI interpretation of video monitoring, ECG, and other features. It looks to me like deep learning will have this on lock very soon. As someone interested in epilepsy/neurophys, it has me wondering what the future holds. Granted this tech will need to be vetted. But it only took neuropace about 5 years. I am seeing a lot of other upcoming automation in other neuro fields, including stroke. I haven't heard a lot of talk about this and I'm wondering your opinions.

It seems like you are already convinced AI will take over despite zero experience with clinical EEG interpretation. Boeing thought it's software was smarter than pilots too, and the company might go bankrupt over that mistake. Persyst is an example of this 'AI' in EEG, and is completely worthless. It cannot tell a sharp wave from a vertex wave, or a seizure from someone eating lunch. EKG is much simpler than EEG with fewer potential diagnoses and less datapoints yet still lacks any halfway reliable computer interpretation.

 

[Postictal Raiden]

From a financial prospective, there’s very little incentive for hospitals to invest in a technology that automates EEGs. The physician’s fee is a small component of the total reimbursement and hardly justifiable to purchase an AI for that purpose. My residency program is part of a hospital system (one of the largest in the US). We lack enough EEG machines to meet the demand. The admins don’t want to purchase more machines because, to my surprise, EEGs aren’t nearly as lucrative as other stuff they rather spend money on.

Another thing to consider is that unlike radiology, EEGs are interpreted by clinicians. Many times by the same clinician that is treating the patient. This serves in reducing false positives and over calling minor abnormalities (or even normal variants). I can easily imagine a world where everyone is on Keppra because of EEG misinterpretation.

Also remember, epilepsy is much much more than EEG interpretation. Seizure remains a clinical diagnosis and the decision to treat patients must be based on clinical suspicion.

 

[Always_Eating]

My research focus is in deep learning of neurologic time series data, specifically ECoG. I have a MD and am a Neurologist but I do this research in the role of the engineer / computer scientist. I work closely with Epileptologists but am not one myself. I do read EEGs clinically though. Epileptologists understand very little about math, computer science, and machine learning. They do not know the difference between a GAN, a CNN, a MLP, or SIREN vs ReLU vs PReLU. Independent of a GUI being handed to them, they will be overwhelmed by the pace of development in deep learning research. Papers that were hot at NeurIPS even 5 years ago would get yawns from attendees today. The other issue is that the SOTA changes month to month, whereas clinical practice changes slowly. With end-to-end ML reads without Epileptologist involvement, all it takes is one malpractice lawyer with a CS degree to say "Why didn't you use Transfer Learning with Fourier Transforms in a Graph CNN per Blah et al. SOTA paper published last week at EMBC? Your algorithm FN rate is 0.001 but their work showed FN rate of 0.0001."

What will likely be the first phase of development is that an end-to-end ML interface will pre-read EEGs and flag segments of interest for the Clinical Epileptologist to focus on during a formal read. I have confirmed with numerous Epileptologists that they are open to this and feel it is the first acceptable paradigm shift. This protects the engineers in that they are not making the final call, as alternatively if there was no MD involved then false negatives could lead to the tech company being sued. This also protects the MD in the same way from not missing things and he/she can rely on the algorithm by choice or read every segment if desired.

I see something similar on the horizon for Radiology, and when pre-reads become standard, it could reduce the number of working Radiologists by 80% or more. A single Radiologist with ML pre-reads will be able to do the work of 3-5 Radiologists. Same thing will probably happen for EEG.

I have also submitted papers for ML in stroke (Computer Vision for detecting focal deficits via Tele encounters), but this is well behind EEG and Rads reads.

Check-out Yannick Roy recent JNE review on DL for EEG analysis: Deep learning-based electroencephalography analysis: a systematic review - IOPscience

Lastly, if you are interested in making a career of ML/DL research, you will have to leave clinical work to focus on research entirely because there are too many smart, hard-working people in this field now with backgrounds in Physics, Applied Math, CS, EE from MIT, Stanford, Berkeley, etc. They dedicate all of their time to this and a Clinician simply cannot keep up unless acting as a collaborator. It seems like if you do not post to Arxiv on Monday then your idea will be posted by someone else on Wednesday same week.

 

[xenotype]

What will likely be the first phase of development is that an end-to-end ML interface will pre-read EEGs and flag segments of interest for the Clinical Epileptologist to focus on during a formal read. I have confirmed with numerous Epileptologists that they are open to this and feel it is the first acceptable paradigm shift. This protects the engineers in that they are not making the final call, as alternatively if there was no MD involved then false negatives could lead to the tech company being sued. This also protects the MD in the same way from not missing things and he/she can rely on the algorithm by choice or read every segment if desired.

This is potentially realistic but I disagree. Automated EKG analysis has not saved anyone any time. With EEG skilled epileptologists already fast forward over segments with a low likelihood of abnormalities AND the EEG has already been prescreened by a tech as you are well aware. None of this labor is really spared by a machine pre-read. Some epileptologists already barely look at anything but the events and the first night of sleep. Some only look at the segments processed by a tech instead of screening the entire EEG. The engineers and the company behind them will NEVER be able to take the liability for final reads. A single mistake can bankrupt a company, as per my Boeing reference. A series of mistakes with a common factor is much more likely with software, and amplifies the financial and regulatory risk exponentially.

Sure software will provide pre-reads. So far the software isn't very good. Perhaps you know of super secret software in the research realm that is way, way better than Persyst. If you can point to super secret software that gets EKG 99% correct to the standard of an electrophysiologist, I'll change my opinion. Automated EKG analysis has existed for a long, long time now and still sucks horribly. EEG is even more complicated for software interpretation AND includes video with immeasurable possibilities.

In general I think there is a massive amount of hubris on part of software engineers in regards to AI and neural networks. The software will get better iteratively, but the amount of time this will take and the amount of legal risk involved is underestimated pretty heavily. Here is >10 years of driverless car development demonstrating only yesterday it's vast limitations. I already mentioned the 737 MAX disaster in which pilot inputs were overriden by software receiving erroneous sensor data killing almost 400 people in two separate but identical crashes. Lawyers and government regulators don't care about the enthusiasm and confidence, and lawyers love the deep pockets that engineering companies have.

 

[DerpyNeuroMD]

I get a kick out of "it's the end of the world! the sky is falling!" type comments. I think it would be incredibly premature to raise any alarm bells considering we've hardly got any artificial intelligence taking over physician duties (unless you count midlevels *ba-dum-tss* 😅). At this point it seems more limited to the research domain -- strengthening and validating clinical indices and whatnot. Anyway.

It seems like you are already convinced AI will take over despite zero experience with clinical EEG interpretation. Boeing thought it's software was smarter than pilots too, and the company might go bankrupt over that mistake. Persyst is an example of this 'AI' in EEG, and is completely worthless. It cannot tell a sharp wave from a vertex wave, or a seizure from someone eating lunch. EKG is much simpler than EEG with fewer potential diagnoses and less [sic] datapoints yet still lacks any halfway reliable computer interpretation.

Hi xenotype. I know you're quite a frequent poster here and I hope not to offend or be argumentative. I'm just starting my career in neurology and intend to visit this forum often! So please allow me to respectfully say I don't mean to put people on the defensive and your reply isn't addressing exactly what I meant. I just meant to open the discussion considering it seems not to have been discussed. AI is growing in ubiquity in medicine. That's not to say I am convinced AI will take over. I'm just speaking from the perspective of an experienced computer scientist and engineer that these problems are apparently growing more trivial for computational analysis.

Persyst probably is worthless. After reading through their publications I see no mention of AI or machine learning. It looks like they're rooted in 2014 tech, so they're probably using some Douglas–Peucker algorithm-type stuff, which I loved to use back then when writing my algorithms for analyzing polysomnography.

I am saying now we have hybrid deep neural networks which can utilize all of the things I mentioned to forecast and/or predict. Implanted and wearable tech can add features like trends in blood gas, K+, Na+, pH, glucose, osmolality, lactate, cortisol, actigraphy, polysomnography, circadian and diurnal data, barometric pressure, moon phase, and etc. to the equation. Also clinical and socioeconomic data -- all of the above have been correlated with seizure. We're talking data cubes of enormous dimensionality. I believe the reality is computers are probably already outperforming humans, but we'll see more practical implementations very shortly.

Bill Gates likes to say: "We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten. Don't let yourself be lulled into inaction."

This is potentially realistic but I disagree. Automated EKG analysis has not saved anyone any time.

I disagree. AEDs have presumably saved many lives.

My research focus is in deep learning of neurologic time series data, specifically ECoG. I have a MD and am a Neurologist but I do this research in the role of the engineer / computer scientist. I work closely with Epileptologists but am not one myself. I do read EEGs clinically though. Epileptologists understand very little about math, computer science, and machine learning. They do not know the difference between a GAN, a CNN, a MLP, or SIREN vs ReLU vs PReLU. Independent of a GUI being handed to them, they will be overwhelmed by the pace of development in deep learning research. Papers that were hot at NeurIPS even 5 years ago would get yawns from attendees today. The other issue is that the SOTA changes month to month, whereas clinical practice changes slowly. With end-to-end ML reads without Epileptologist involvement, all it takes is one malpractice lawyer with a CS degree to say "Why didn't you use Transfer Learning with Fourier Transforms in a Graph CNN per Blah et al. SOTA paper published last week at EMBC? Your algorithm FN rate is 0.001 but their work showed FN rate of 0.0001."

What will likely be the first phase of development is that an end-to-end ML interface will pre-read EEGs and flag segments of interest for the Clinical Epileptologist to focus on during a formal read. I have confirmed with numerous Epileptologists that they are open to this and feel it is the first acceptable paradigm shift. This protects the engineers in that they are not making the final call, as alternatively if there was no MD involved then false negatives could lead to the tech company being sued. This also protects the MD in the same way from not missing things and he/she can rely on the algorithm by choice or read every segment if desired.

I see something similar on the horizon for Radiology, and when pre-reads become standard, it could reduce the number of working Radiologists by 80% or more. A single Radiologist with ML pre-reads will be able to do the work of 3-5 Radiologists. Same thing will probably happen for EEG.

I have also submitted papers for ML in stroke (Computer Vision for detecting focal deficits via Tele encounters), but this is well behind EEG and Rads reads.

Check-out Yannick Roy recent JNE review on DL for EEG analysis: Deep learning-based electroencephalography analysis: a systematic review - IOPscience

Lastly, if you are interested in making a career of ML/DL research, you will have to leave clinical work to focus on research entirely because there are too many smart, hard-working people in this field now with backgrounds in Physics, Applied Math, CS, EE from MIT, Stanford, Berkeley, etc. They dedicate all of their time to this and a Clinician simply cannot keep up unless acting as a collaborator. It seems like if you do not post to Arxiv on Monday then your idea will be posted by someone else on Wednesday same week.

Hey me too! As I said above, I'm a computer scientist and engineer, and much of my research has involved machine learning in various domains from waveform analysis to computer vision.

I see you're focused on time series data. I'm wondering what the future holds for accessible hybrid NNs. (I think Keras has something?) Anyway I guess you're probably comfortable with LSTMs or other RNNs, but I'd like to see some studies involving RNNs + CNNs. I mean, I'd like to see models of cube dimensions trained with different algorithms combined to predict the dependent variable(s), if that makes sense. I haven't looked into this much though. But I mean, Google search's spelling model was trained with 680 million features. The tech is getting there.

Yeah! To clarify, I've stated before elsewhere -- and agree with your assessment that -- the role of AI in medicine (especially in rads where it's the talk of the town) will be more of a utility, especially at first. I.e. radiologists would possibly be able to read more images because the models would highlight the important parts, yet keep the radiologist responsible for accepting/declining the prediction (which would further train the model!). I think sleep and eeg are much simpler and less noisy problems.

Agreed that the PhDs working on this stuff have a good grip on things. And I would also agree that clinician-scientists are a critical part of the process.

Also a review article from 2019 seems quite dated in this field I would say, right? Pubmed shows twice as many papers with keywords "machine learning" in 2020 as there were in 2018. The growth in knowledge looks exponential

 

[xenotype]

Hi xenotype. I know you're quite a frequent poster here and I hope not to offend or be argumentative. I'm just starting my career in neurology and intend to visit this forum often! So please allow me to respectfully say I don't mean to put people on the defensive and your reply isn't addressing exactly what I meant. I just meant to open the discussion considering it seems not to have been discussed. AI is growing in ubiquity in medicine. That's not to say I am convinced AI will take over. I'm just speaking from the perspective of an experienced computer scientist and engineer that these problems are apparently growing more trivial for computational analysis.

Persyst probably is worthless. After reading through their publications I see no mention of AI or machine learning. It looks like they're rooted in 2014 tech, so they're probably using some Douglas–Peucker algorithm-type stuff, which I loved to use back then when writing my algorithms for analyzing polysomnography.

I am saying now we have hybrid deep neural networks which can utilize all of the things I mentioned to forecast and/or predict. Implanted and wearable tech can add features like trends in blood gas, K+, Na+, pH, glucose, osmolality, lactate, cortisol, actigraphy, polysomnography, circadian and diurnal data, barometric pressure, moon phase, and etc. to the equation. Also clinical and socioeconomic data -- all of the above have been correlated with seizure. We're talking data cubes of enormous dimensionality. I believe the reality is computers are probably already outperforming humans, but we'll see more practical implementations very shortly.

Bill Gates likes to say: "We always overestimate the change that will occur in the next two years and underestimate the change that will occur in the next ten. Don't let yourself be lulled into inaction."

I disagree. AEDs have presumably saved many lives.

The level of education required to interpret the EKG rhythms that an AED interprets is EMT-basic. That is far from diagnosing a subtle posterior MI, a brugada, or a complex AV block. If the software was great, EM physicians could look at the automated interpretation and ignore the actual EKG unless there was a significant finding. In practice, the automated analysis is ignored completely regardless of what it says, and the EKG is rapidly analyzed manually. If EEG is trivial for analysis, EKG should be significantly easier and should have already been done to perfection by now.

I'll use another example since you ignored my earlier ones. Self driving cars cost between $200-400k, but the extreme cost is having to help them every time they get stuck, which for self driving companies requires nearly 1:1 supervision even if there isn't a driver in the actual car. The other alternative is being Tesla with cheaper sensors and more basic software, proclaiming your cars are safe, and incurring high profile NTSB investigations and lawsuits plummeting your stock even when you try to dump the liability on the driver. Sure the software will improve as they get more data. How much will it improve and how fast? Right now there is zero profit in 'autonomous vehicles' after >10 years of development. That isn't to say the software is useless- automatic breaking, lane departure, advanced cruise controls are widely available and obviously useful. The truly autonomous vehicle that needs no babysitting is still decades away, and a single mistake can sink one of these companies.

Call me a luddite if you want, but I don't believe at all the epileptologist/neurophysiologist will be chopped out of EEG interpretation in my career. The software might reduce the need for tech processing and help flag segments for review, and certainly could be most useful for automated alerts for ICU studies once the false positives get down to a reasonable rate.

 

[Always_Eating]

Automated EKG analysis has existed for a long, long time now and still sucks horribly. EEG is even more complicated for software interpretation AND includes video with immeasurable possibilities.

Xenotype, I think we are on different pages in terms of talking about the current state of ML for interpretation for ECG/EEG. When you talk about the difficulties in ECG, you are talking about an end-to-end read which becomes a multi-label classification problem. This means the ECG needs to detect rate, rhythm, MI, blocks, etc. In a very basic form, you could think about the model needing to learn 4+ different sub-models to label each of these things. When talking about EEG, there are papers with 99% accuracy on seizure-detection in isolation (A Deep Learning Approach for Automatic Seizure Detection in Children With Epilepsy), which is a binary classification problem: seizure vs not seizure. The multi-label classification problem of full seizure interpretation (PDR quality, seizures, state changes, dominant frequency, response to hyperventilation, and so on) will come in time, but yes we are a ways away from that, including in the NLP output of a full report that is EMR ready. I do not want to get too in the weeds on ML, but there is a big difference between our claim that deep learning is approaching Epileptologist-level accuracy in seizure detection, vs what I think you are thinking of, which is deep learning replacing Epileptologists entirely. I am not saying the latter.

But if you are an Epileptologist you can easily see step one. You actually mentioned it yourself: removing the techs from the pre-read situation. When I was a resident we had a very large inpatient EMU. There were 3 or more techs on 24/7 looking at dozens of patients' EEGs on screens in front of them. I do not know what they were paid but I am certain an open-source EEG reading algorithm for seizure screening for the reading Epileptologist would be less expensive.

 

[xenotype]

Xenotype, I think we are on different pages in terms of talking about the current state of ML for interpretation for ECG/EEG. When you talk about the difficulties in ECG, you are talking about an end-to-end read which becomes a multi-label classification problem. This means the ECG needs to detect rate, rhythm, MI, blocks, etc. In a very basic form, you could think about the model needing to learn 4+ different sub-models to label each of these things. When talking about EEG, there are papers with 99% accuracy on seizure-detection in isolation (A Deep Learning Approach for Automatic Seizure Detection in Children With Epilepsy), which is a binary classification problem: seizure vs not seizure. The multi-label classification problem of full seizure interpretation (PDR quality, seizures, state changes, dominant frequency, response to hyperventilation, and so on) will come in time, but yes we are a ways away from that, including in the NLP output of a full report that is EMR ready. I do not want to get too in the weeds on ML, but there is a big difference between our claim that deep learning is approaching Epileptologist-level accuracy in seizure detection, vs what I think you are thinking of, which is deep learning replacing Epileptologists entirely. I am not saying the latter.

But if you are an Epileptologist you can easily see step one. You actually mentioned it yourself: removing the techs from the pre-read situation. When I was a resident we had a very large inpatient EMU. There were 3 or more techs on 24/7 looking at dozens of patients' EEGs on screens in front of them. I do not know what they were paid but I am certain an open-source EEG reading algorithm for seizure screening for the reading Epileptologist would be less expensive.

Electrographic seizure detection in isolation is simple, and Persyst already does this with semi-reasonable accuracy, but has a large, large amount of false positives. Our center is probably 10-20% FLE, EEG negative. Good luck. Sure, you can potentially save some money in eliminating the need for tech screening. 'End to end' as you describe is the only thing that will cut the neurologist out of the loop- it is the same with 'autonomous vehicles'. Either it can function completely independently of human input or it cannot. The economic variables in this equation do not care about 'multi-label problems'. This is why your system cannot replace an epileptologist, and won't be able to for a long time. Even a taxi driver cannot be replaced by AI without a van following it to fix it when it gets stuck. IBM Watson thought they would be god's computerized gift to medicine, but the technology and large amount of engineering and funding behind it has essentially been totally worthless.

 

[DerpyNeuroMD]

Electrographic seizure detection in isolation is simple, and Persyst already does this with semi-reasonable accuracy, but has a large, large amount of false positives. Our center is probably 10-20% FLE, EEG negative. Good luck. Sure, you can potentially save some money in eliminating the need for tech screening. 'End to end' as you describe is the only thing that will cut the neurologist out of the loop- it is the same with 'autonomous vehicles'. Either it can function completely independently of human input or it cannot. The economic variables in this equation do not care about 'multi-label problems'. This is why your system cannot replace an epileptologist, and won't be able to for a long time. Even a taxi driver cannot be replaced by AI without a van following it to fix it when it gets stuck. IBM Watson thought they would be god's computerized gift to medicine, but the technology and large amount of engineering and funding behind it has essentially been totally worthless.

Not sure why youre arguing this as a layman. What's your background in machine learning or data science? I know you've published 500+ papers before residency but I'd love to see some on the related topic

 

[xenotype]

Not sure why youre arguing this as a layman. What's your background in machine learning or data science?

What is your background on clinical EEG interpretation? Why are you questioning my background instead of responding to my comment? You made this thread in a Neurology forum, not a 'data science' forum.

 

[DerpyNeuroMD]

What is your background on clinical EEG interpretation? Why are you questioning my background instead of responding to my comment? You made this thread in a Neurology forum, not a 'data science' forum.

Which comment? The self-driving car one? A handful of waveforms is nowhere near a comparable problem to a self-driving car (????). The fact that a problem of such incredible magnitude is actually near solved should give you at least a glimmer of cognizance that a solution for such a relatively simple one may be nearing implementation.

I've given a fair overview of the current state of ML in seizure after having done hundreds of hours of lit search for my paper on this very subject. You're belligerently arguing some unrelated left-field points. I don't know what you want from me other than saying that you're right, but you aren't. Go do the research yourself: ((machine[Title]) AND (learning[Title])) AND (eeg[Title]) - Search Results - PubMed

 

[xenotype]

Which comment? The self-driving car one? A handful of waveforms is nowhere near a comparable problem to a self-driving car (????). The fact that a problem of such incredible magnitude is actually near solved should give you at least a glimmer of cognizance that a solution for such a relatively simple one may be nearing implementation.

I've given a fair overview of the current state of ML in seizure after having done hundreds of hours of lit search for my paper on this very subject. You're belligerently arguing some unrelated left-field points. I don't know what you want from me other than saying that you're right, but you aren't. Go do the research yourself: ((machine[Title]) AND (learning[Title])) AND (eeg[Title]) - Search Results - PubMed

You made this thread with a statement that the sky is falling and epileptologists will soon be obsolete. I don't think the research shows that at all, and I don't think you are right. You don't care about my opinion, or potentially relevant analogies from other areas of software/AI replacing human judgement, and that is fine.

 

[deathmerchant]

Yea I think automation for EEG is good as a tool for developing AI and exploring its capabilities. But the few problems I see there are.

1. As mentioned above, there's not enough money in EEGs anymore. The number of EEGs and reimbursement are really low. For eg, I am the only person who reads EEGs in our community hospital and the RVUs are under 7% of our total RVUs. There's isn't much incentive to change the system for admins and for physicians.

2. Routine EEG has very very low sensitivity, and honestly outside of specialized epilepsy clinics, over 90% of times it doesn't change management. We still rely way more on clinicals than EEG.

3. By the time AI gets advanced enough and/or accepted widely, we probably would have found new aspects of it and AI will need to be redirected again (although that could be done fast). Furthermore, we need neurologists to keep doing research and clinical correlations to find more indications/uses/variants of EEG.

Also as mentioned above, I haven't even seen EKG recording being automated. EEG is way way more noisy, variable and complex.
Pre-read of EEG followed by neurologist reads seems reasonable and feasible. Although most techs do a pre-read and we will always need them. So wouldn't make sense to have AI read it. Unless its extremely cheap.

 

[xenotype]

I hesitate to revive a dead topic but I think this new STAT article is prescient on this issue, and STAT seems to really be the only one taking a critical well researched view of AI in medicine. Many of the AI models are garbage with way too much confidence in them based on flawed methodology and oversimplifications, and STAT backs this up with data and interviews from the radiology world including significant and accurate criticism of an article published in Nature. I absolutely believe it applies to EEG.

 

[Always_Eating]

I hesitate to revive a dead topic but I think this new STAT article is prescient on this issue, and STAT seems to really be the only one taking a critical well researched view of AI in medicine. Many of the AI models are garbage with way too much confidence in them based on flawed methodology and oversimplifications, and STAT backs this up with data and interviews from the radiology world including significant and accurate criticism of an article published in Nature. I absolutely believe it applies to EEG.

I suggest you read the original Hansen article at On instabilities of deep learning in image reconstruction and the potential costs of AI before making such sweeping claims.

As a physician and AI researcher, I find that the loudest critics from the clinical side are those that have not taken the time to gain an understanding of AI applications and limitations.

Hansen's group is specifically addressing a newer and less mature application of deep learning, which is image reconstruction for inverse problems.

In a very basic form, you can think of this as an AI model producing an estimation of T2, SWI, DWI, and ADC sequences from T1 sequence data alone. And yes, this is a newer application of AI that is still in relative infancy.

There is a big difference between the classification problem in EEG and inverse problems in general radiology.

 

[xenotype]

As a physician and AI researcher, I find that the loudest critics from the clinical side are those that have not taken the time to gain an understanding of AI applications and limitations.

You didn't address or comment on the concerns raised in the STAT article at all. Variability and artifact in real world samples is absolutely the concern.

 

[dodobirddh]

AI may become a tool at the hands of neurologist in the near future..
FDA approved AI tools like "Rapid LVO"...promises to speed up the acute stroke management pathway.
Rapid LVO
I have also found AI useful in looking up quick differential diagnosis on mobile devices... "Neurology Pro"
Neurology Pro

The AI tools have been backed by multicenter studies showing improved performance by using the AI tools..
research paper

AI may not however completely replace the neurologist in the foreseeable future... my two cents!

 

[xenotype]

AI may become a tool at the hands of neurologist in the near future..
FDA approved AI tools like "Rapid LVO"...promises to speed up the acute stroke management pathway.
Rapid LVO
I have also found AI useful in looking up quick differential diagnosis on mobile devices... "Neurology Pro"
Neurology Pro

The AI tools have been backed by multicenter studies showing improved performance by using the AI tools..
research paper

AI may not however completely replace the neurologist in the foreseeable future... my two cents!

Yes, viz.ai and RAPID are useful tools for LVO evaluation, but as is key to most of the involvement of AI in medicine they can in no way replace clinical judgement. In the case of the AI tools for stroke, they don't provide a diagnosis, moreso speed image processing to bypass PACS for perfusion data. It doesn't guarantee there is an LVO or not and has many limitations. AI tools can certainly help interpret procedural and clinical data more quickly (and I have no disagreement they may help process EEG recordings quicker in the future). My contention is they will not accurately spit out a final, actionable diagnosis on their own reliably enough to avoid significant work input from the physician anytime soon. Once EKG has been solved I may change my opinion, as this should be the absolute simplest medical test to automate completely with the least amount of data points, least amount of artifact, and smallest differential diagnosis, yet it has not been done despite decades long attempts.