Finding Biomarkers in Huntington’s Disease

The applications of AI technology are rapidly expanding as Canary Speech identifies biomarkers through voice. Canary’s technology can be trained to detect conditions or illnesses that are traditionally difficult to identify, or those where early identification is crucial to treatment.

These capabilities prompted a partnership between Beth Israel Deaconess Medical Center (Harvard Medical School’s teaching hospital) and Canary Speech. Together, they successfully detected speech changes in patients of Huntington’s disease (HD), a genetic disorder that progressively breaks-down brain cells. The symptoms of HD vary in both degree and presentation. While involuntary or impaired movements are the most obvious symptoms of HD, some individuals experience cognitive and psychiatric disorders which are much less overt. Approximately 30,000 individuals have Huntington’s disease in the US.

Traditional solutions may do too little, too late

Currently, healthcare professionals gauge speech motor impairment by rating their patients on a scale of 1-4. This presents two problems; first, this method is highly subjective to the individual conducting the examination. Second, fluctuations in speech patterns are very subtle in early stages of the disease and are therefore difficult to identify until the disease has progressed to a more advanced stage.

Even more troubling, motor impairment is only one symptom of HD. Psychiatric and cognitive disorders change speech patterns, but recognizing these changes is much more difficult than with motor impairment. This can be especially detrimental for undiagnosed individuals with HD. Psychiatric and cognitive decline may present as depression, OCD, bipolar disorder, and slowness in processing thoughts leading to misdiagnosis and inappropriate treatment.

AI can identify subtle changes earlier

AI technology is powerful. Not only can it identify changes in speech earlier than the human ear, it can identify changes in individuals affected psychiatrically and cognitively as well.

At the annual meeting of the Huntington Study Group in November, Beth Israel presented findings of its joint study with Canary Speech, “Audio Analysis of Acoustic and Linguistic Features in Huntington’s Disease.” In this trial, Canary Speech’s model was successfully trained to identify vocal biomarkers of HD.

Canary’s biomarker data enabled the analysis which identified more than 1,000 features of speech differentiating healthy patients from HD patients. These features include speech dynamics, duration per word, words per second, bandwidth, and contrast, along with many others.

Four charts indicating the difference in means between healthy patients versus patients with Huntington's disease according to vocal feature.

Sierra, L.A, Ullman, C.J, Hildebrand, K, Dierker, J.S, Mwangi, M, O’Connell, H, Frank, S.A, Laganiere, S. 2022. Audio Analysis of Acoustic and Linguistic Features in Huntington’s Disease. Annual Huntington Study Group, 3-5 November, Tampa, FL.

AI and Huntington’s, moving forward

Canary Speech’s technology can detect sensitive speech changes related to HD. Now that indicative vocal features have been identified, machine learning will be able to generate better models to detect early changes in HD.

In the future, we see Canary Speech’s technology widely used as the gold standard to screen and monitor patients experiencing psychiatric and cognitive decline. With many symptoms that don’t immediately point to HD, Canary’s technology can rule out expensive and invasive tests for patients or provide healthcare professionals a real-time indication that further HD testing is necessary. 

The results of this training are exciting and additional trials are scheduled to continue honing the HD model. Download the Beth Israel – Canary poster here.