Waveform-based analysis of head tremor using a marker-less tracking algorithm with 2D-video

Jung, it's great to have you on the podcast again today. You've become a regular. And Dr. Bj, it's really an honor to have you here today. So my first question for you both and either of you can chime in on this, I wanted to ask you how did you come up with the idea of doing this study?

Was this something that you had been thinking about for a while, or did you come across the new technology, the video-based algorithm, and thought that this might be a good use for it? Or were you inspired by a particular patient encounter?

Dr. Beomseok Jeon: This problem all start when I joined Dystonia Tremor Working Group, which was organized by MDS which was to specifically work on this problem dystonic tremor.

This working group was headed by Hyder Jinnah and had [00:02:00] many experts like Alberto Albanese, Alfonso Fasano, Sanjay Pandey, Rick Helmich, Marie Vidailhet, Marina Tjissen, Aasef Shaikh, Roger Elble, and Victor Fung. And one of the member was Mark Hallett, who passed away several weeks ago, who made a great contribution to this work making many good comments and sending patients videos to us.

We really wanna thank him very much.

Dr. Mitra Afshari: Absolutely. I think we're all missing Dr. Hallett and we're thinking about him. So it sounds like there was a lot of work that went into the idea behind this project and a lot of work on behalf of multiple powerhouses when it comes to dystonia, as part of the dystonia working group. And so in this study, you aim to find a more objective approach to [00:03:00] tremor analysis and cervical dystonia and essential tremor patients who have head tremor. Anecdotally, I think we all have been in that position where we've questioned our ability to really differentiate a dystonic head tremor from a head tremor that's seen in essential tremor. We know, dystonic tremors seem to be less rhythmic. They seem to be less sinusoidal, they're more jerky. They're less of that kind of pure, true tremor that we think of, and while you are also looking for additional clues like underlying tonic posturing, there are definitely some cases that can be very challenging. So you asked the simple question, let's take what we think we know about the rhythmicity and the sinusoidality of these head tremors and let's put them to the test. So you use simple computer vision techniques to analyze 2D video recordings of patients with head tremors, specifically [00:04:00] tracking the movement of the tip of the nose, and you looked at patients prospectively and retrospectively through archive videos.

As you mentioned, Dr. Bj, you specifically partnered with the well-known Dystonia Coalition Group that's led by Buzz Jinnah at Emory University in Atlanta in the US and you really leveraged their archive of cervical dystonia patients of those videos. The patient specifically with head tremor and you also prospectively enrolled patients at your university and even took advantage of some videos of ET patients who were undergoing MRI focused ultrasound, if I'm correct, for their hand tremors. You were able to also confirm your findings from these markerless video-based waveform analysis. You confirmed that these findings were in agreement with physical sensor-based measurements by correlating the data with a gyroscope that was placed the patient's head. So Dr. [00:05:00] BJ or Dr. Shin, could you briefly describe your methods in this study, and ultimately, what did you end up doing? Could you paint a picture in our listener's mind of how all of this worked out? 

Dr. Jung Hwan Shin: Yeah. Thank you very much for the question. It's great to see you very soon, Mitra. I've been interested in video based analysis of movements in movement disorders including Parkinson's disease and actually my research interests focus on this video based analysis of bradykinesia, postures and gait especially, I'm quite focused on doing the gait analysis with the two dimensional videos. But it's quite interesting that I have been interested in this video-based analysis, which all started from my PhD era where I did the basic research using the mouse models. Back in the days when I did the PhD, I had trouble tracing the mouse wearing like endoscopes.

So there were like [00:06:00] lines attached to the mouse head and the conventional method confused those lines attached to the mouse heads, like endoscopy with the tail of the mouse. So we have to come up with the different method to analyze the trajectory of the mouse. And we found out that computer vision techniques could enable the tracking of the mouse without markers attached to the actual mouse. So, my PhD program, I used a lot of those algorithms to trace position of the mouse. And after I came back to the clinic, I found that we have a lot of videos. We have a lot of videos of the tremors, gait, bradykinesia, so why not apply this algorithm to the patients? So that's all it began.

So I did many studies even on many phenomenology of the patient with under supervision of the professor Bj. And this project started [00:07:00] from the lab meetings where BJ shared the discussions on dystonia tremor groups that he mentioned. Although there are a clear definition of the tremor and dystonia, but when it comes to the actual phenomenology of the patients there could be a lot of disagreement because whether we perceive rhythmicity and sinusoidal are quite subjective. So, Bj mentioned why not measure those sinusoidal and rhythmicity with the video-based algorithm because it reflects what because it's a video analysis. Okay. So he mentioned that it would be very beneficial. So this project started with the hypothesis driven way, not the data-driven way. So it was very clear to measure sinusoidality and rhythmicity, not the whole comprehensive feature of this tremor. So I think that was very clear and lead the way which made the study [00:08:00] more feasible and efficient in some way. So that's why we measured only the nose tip of the head tremors because if we were to analyze the many comprehensive features of the head tremors, we might have to use three dimensional method because head tremors goes in the three dimensional way. But our goal was to measure rhythmicity and sinusoidality. So we didn't have to measure all of those features, but we need to derive the waveform of the patient.

Only tracking the nose tip of the head tremors we could analyze how rhythmic or how sinusoidal their tremor waveforms are. And we had to develop our own parameters to measure the rhythmicity and sinusoidality , which have been like trial and error. And we came up with the best parameters that could [00:09:00] reflect the rhythmicity and sinusoidality, which many of our tremor and dystonia experts agree the co-authors of our papers. This study we wanted to validate whether rhythmicity and sinusoidality parameters that we calculated are correct. So we validated with our internal cohorts using the gyroscope. We saw that those parameters matched and from the start, we used like the high frequency, no high frame rate videos, 240 frame per second videos because we wanted to make sure that we couldn't have like the analyzing effect because we knew video record the patient with tremors with the low frame rate. Then you the analysis mislead to the different frequencies of the actual tremor. So we used 240 frame per second videos, but we [00:10:00] eventually found that even with the 30 frame per second when we down sample those videos, it still showed a good agreement with the gyroscope. So, we wanted to test this algorithm to the many videos, not from our own center, but videos from around the world. And Dystonia Coalition gave a lot of support of sharing those videos. Not only the cervical dystonia patients, but also the essential tremor patients. And that's how we showed the distribution of the rhythmicity and sinusoidality in cervical dystonia and essential tremor patients. So basically the outcome, the main result of this study is that we showed the distribution of those cervical dystonia essential tremor in terms of rhythmicity and sinusoidality, which was objectively measured by the video analysis.

Dr. Mitra Afshari: Gotcha. So you took something that you had [00:11:00] been applying to your animal models and you essentially applied them to humans.

And you were able to test out a simple hypothesis, and really confirmed that hypothesis. So it was definitely a creative approach. And so you got into some of the results of the paper. So I guess everyone wants to know, are dystonic tremors in cervical dystonia, are they indeed less rhythmic and less sinusoidal? And did it seem like your results fit with that hypothesis?

Dr. Beomseok Jeon: Of course, yes. Head movements in cervical dystonia are worse, less rhythmic and less sinusoidal in many cases. That's why some people in our group did not want to use the term dystonic tremors because tremors should be rhythmic and sinusoidal. Some people in our group did not agree with the term dystonic tremors in cervical [00:12:00] dystonia, but there were some people who had movements in cervical dystonia, who had cervical dystonia, which was pretty rhythmic and pretty sinusoidal. He chose in the range of essential tremor that was the beginning of the problem, that needs to be solved in future.

Dr. Mitra Afshari: Fantastic. And did you find that the tremor amplitude specifically influenced your results? I think that was one of the variables that you looked at specifically. Could you review those findings with us?

Dr. Jung Hwan Shin: Yes, indeed. It had to do with the amplitude and we had a lot of videos that we analyzed and we found that there are patients with the large amplitude. There were patients with the small amplitude and we found that those sinusoidality and rhythmicity index correlated with the tremor amplitude.

[00:13:00] If the tremor was amplitude was large, they tend to be rhythmic and sinusoidal. But in the lower amplitude group, they tend to be less sinusoidal and less rhythmic. But the limitation of our study was that we couldn't objectively measured amplitude of the tremor because we used the two dimensional video.

So the distance between the head and the camera was different among individuals. If we could control the distance, maybe we could objectively derive the amplitude. But in this study we couldn't do that. So we indirectly showed the tremor amplitude correlated with the rhythmicity and sinusoidality index based on the tetra scale and the ambiguous tremor amplitude derive from the patient. But this has been studied in the previous study, including many researchers including Dr. Aasef showing that indeed [00:14:00] tremor are correlated with their sinusoidality and rhythmicity of the tremors.

Dr. Mitra Afshari: So even though there was some limitation as to how you derive that data, what you did find did seem to fit with previous studies with respect to the tremor amplitude. So there's definitely something there. One interesting finding was that in the cohort for the ET patients, while there was higher rhythmicity in sinusoidality which was, more indicative that it was essential tremor. There was less accuracy between the quote unquote video-based diagnosis and the actual clinical diagnosis. So that diagnostic accuracy was about 64% when it came to ET compared to about 82% when it came to cervical dystonia head tremor. What do you make of this finding?

Dr. Beomseok Jeon: Again cervical dystonia has less rhythmic, less sinusoidal head [00:15:00] movements that was very clear that differentiates cervical dystonia from other tremor disorders. But some patients with dystonia had a rather rhythmic, rather sinusoidal movements, features close to essential tremor. That's why just based on rhythmicity and sinusoidality, they cannot reliably differentiate ET versus cervical dystonia.

That's the problem. That has been the problem that whether cervical dystonia patients can have rhythmic, sinusoidal head movements that can be called tremors.

Dr. Mitra Afshari: Absolutely. And that's the challenge that brings us back to why these investigations are done. So putting all of this together, what do you think is the take home message from this investigation? It seems like ET and cervical dystonia head [00:16:00] tremors are definitely all on a spectrum, right? How would you summarize the take home message from your study today?

Dr. Jung Hwan Shin: Okay. So as Professor BJ mentioned, we found variability in heterogeneity of the cervical dystonia with tremors and also within the essential tremor patient as well. I think this study offers that in real world where clinical phenomenology could be like subjectively measured and it's quite difficult to be agreed on even on experts. So there were quite interesting paper by Elan Louis that using the videos of five patients and they asked seven MDS experts to diagnose is this essential tremor or dystonic jerky dystonia. And [00:17:00] all the experts diagnosed those patients within different phenomenology. Some expert said all of those patients are jerky dystonia and some experts said they're all essential tremor patients. And other experts were in between. So we can see in the real world it would be very difficult.

Even we have the definition of each phenomenology would be very difficult. So I think in that terms using the objective measures could give the clue, hint to solve this discussions and controversies. So I think it would be very difficult to just dive into the certain phenomenology based on only using rhythmicity and sinusoidality. We could have the objective measure, but with this we could use other clinical hint like the sensory trick or postures together we could make the more [00:18:00] agreeable to a better description of the phenomenology. For me, I think is to utilize the most recent advanced digital technology with the description of the symptoms could be very helpful in many ways. We could have like distribution as shown in our results table. And we could start discussions from there. So that's what we wanted to propose to the field in terms of this tremor. One of the things that we have done, relatively novel in this study that we derived the sinusoidality Index. The rhythmicity index have been studied and there were many parameters that were developed over the years, but the sinusoidality index was newly proposed in this study. And the reason why we could do it was that we used the video based analysis. We could actually have the wave form of the tremor. And that's [00:19:00] why we could derive the sinusoidality index, which stands for the two fundamental definition of tremor. It should be a rhythmic and sinusoidal and using the recent developed digital technology, I think those approach could be applied to many other phenomenology and controversial parts in the movement disorder as well.

Dr. Mitra Afshari: Absolutely. I think you know these 2D video-based analyses, while they're very simple, they really are quite a powerful tool to add to your toolbox when you're making diagnoses and ultimately so we can make the most accurate diagnoses and develop more targeted and effective therapies. In this investigation, you were only looking at the tip of the nose . But you alluded to this earlier that looking to the future, you could potentially derive other measures even from those 2D based videos. Is that correct? Other movements or [00:20:00] axes that you. That could offer additional information to lend to a diagnosis. Is this something that your group is working on right now or thinking about potentially investigating in the future with the dystonia study group.

Dr. Jung Hwan Shin: Definitely. As we mentioned, we only focused on the tip of the nose. The main reason was that we wanted to see the sinusoidality and rhythmicity of the, head tremors. There are a lot of other models with the video based analysis that could model the three dimensional feature of the head and use it to assess like the turning angles in terms of the head tilt, head rotation. So on. So there are a lot of comprehensive way to analyze the dystonic tremors in the head or even in the hand. So we are very interested in moving this project forward, using those comprehensive approaches and hope to develop more nuanced and better algorithm in the [00:21:00] future in describing those complex phenomenology.

Dr. Mitra Afshari: Thank you so much Dr. Shin, Dr. BJ. This really was an excellent discussion today. We wanna thank your entire team for working on this project, the entire dystonia study group for all the work that you've been doing, and we wanna thank you for taking the time out of your busy schedules to chat with us. I'm sure it's not easy kind of conveying the methods of this research investigation in this forum, but you both did a fantastic job. So wanna thank you and until next time, bye-bye.

Dr. Beomseok Jeon: Bye.

Dr. Jung Hwan Shin: Thank you. [00:22:00] [00:23:00]