Diagnosing a speech motor disorder is very challenging right now for the field.
We have paradigms that were developed about 30 or 40 years ago, and we haven’t really changed much as a field. That’s because in many of the speech disorders, the symptoms overlap. It’s hard to really get at the heart of what’s happening. Is it motoric? Is it language? Is it cognition? So this is an ongoing struggle for our field: improving the diagnostic accuracy.
I think that’s essential because you want to make sure that you’re treating the right problem. That’s only as good as your diagnostic accuracy.
Our current grant is looking at what they call bulbar decline, which is decline of the function of the head and neck muscles. So with amyotrophic lateral sclerosis, ALS, the motor neurons — for unknown reasons — start to deteriorate. Then an individual will lose control over their body or their head and neck. Neurologists right now diagnose the disease based on the symptoms. There is no biomarker for the disease. There is the genetic variant that’s known for about 15%. The remainder are idiopathic, so there’s no known cause. Neurologists and speech-language pathologists are instrumental to the diagnosis of the disease. And it can take about 18 months to get diagnosed because of the long progression of the disease, and because many of the symptoms are very subtle and difficult for a clinician to identify by eye.
We have a speech science laboratory where we can take fine grain measurements of the voice, the respiratory system, and also the tongue and lip and jaw movements. And we’re trying to leverage that data and that information to state, if a patient came in off the street, could you detect that they had an abnormality somewhere in the speech system based on very quantitative benchmarks. We’re working on protocols now to identify the important measures that could potentially be used by diagnosticians. Not only is it important for assessment, but it’s important for predicting how fast a person will decline.
And another thing that will come from this work is, because we’re focused so much on quantifying the behavior and identifying sensitive diagnostic variables, we can then use those variables in the future for drug trials to determine if there are subtle changes that can’t be detected by clinicians. Right now there is no effective drug for the disease, but there are ongoing drug trials. The outcome measures to see if the drugs are improving speech or swallowing are very limited. So, we’re trying to bring a higher level of diagnostic sensitivity to that area as well.
In this new effort, we’re using machine learning to recognize the movements of the mouth in terms of words, phrases, and phonemes when someone is silently producing speech. We feed the positional data — the X, Y, and Z data that we’re recording from the sensors that we have on the tongue, the lips, and the jaw — into these algorithms and they’ll say, oh the person is saying “hello” or “My name is Jordan” and so on.
When we first started this there really wasn’t precedent for it, and we thought it was rather naive, but it actually worked. We’re getting like 98% accuracy. So we’re shocked that there’s enough information just based on the tongue, lip, and jaw movement to extract some speech or recognize some speech.
So now we’re scaling that up and trying that in disordered populations, saying that if the information is now deteriorating in terms of how much you’re conveying with your articulators, how well can these algorithms do? Well, if they can do well, that means the technology could potentially be used by individuals who have motor speech impairments, and it can be an alternative to producing intelligible speech.
We’ll never replace clinicians. The diagnostic process is so complicated. But these tools sure can help a lot. I think it can transform the way clinicians think. I just was attending a talk yesterday where they were talking about how long it takes to train a student to interpret the modified barium studies. You have the radiological studies of swallowing, and you have to watch about a thousand of those before you can become an expert.
I think we can shorten that time. Rather than make individuals expert systems, and have their own view of the diagnostic process, if we can give them better data, then their process can be a lot more accurate and efficient.
The technology is going in very exciting directions. As a field we’re maturing rapidly here, and the science is really scaling up. Now we have investigators studying the genetic basis of the various disorders. We’re doing a lot of neuroimaging work. And we’re doing technology development. I think in a matter of 10 to 15 years, we’re going to be able to reconcile all these various levels of analyses, which is a great challenge. Right now we all kind of work in our areas of specialty, but I think the writing is on the wall that things are going to eventually improve for our patients, given our ability to understand the biological basis of their various disorders, and improving our assessments and treatments.
That’s on the basic, knowledge side of the equation, but I also see that there is a large number of investigators who are interested in treatment studies, and I think we really need this.
It’ll be really fun over the next decade to see where that research goes, and all the improvements we can make in how we deliver services to our patients.