Voice-based tool may track Huntington’s disease progression
Study: Speech index score could also potentially assess treatment response
Written by |
A speech index score built from smartphone voice recordings tracked the severity of Huntington’s disease and matched closely with standard clinical assessments and brain scan measurements, according to a new study from China.
The index score, derived from measures of loudness variation, mispronunciation rate, pitch strength, and speech timing, allowed researchers to distinguish people carrying a Huntington’s-causing mutation from healthy people with high accuracy. Scores also increased (worsened) with more advanced Huntington’s disease stages.
“These findings support the validity of the Speech Index as a sensitive, noninvasive, and scalable tool for monitoring disease progression and potentially assessing treatment response in HD [Huntington’s disease],” researchers wrote, emphasizing, however, that the tool needs testing in larger, more diverse groups before it is used more widely.
The study, “A machine learning-derived speech index as a biomarker for Huntington’s disease severity,” was published in the Journal of Neurology.
Speech difficulties a common feature of Huntington’s
Huntington’s is an inherited brain disorder caused by an expansion of a repeated DNA sequence in the HTT gene that leads to a combination of motor, cognitive, and psychiatric symptoms.
Current tools for tracking the disease, such as the Unified Huntington’s Disease Rating Scale (UHDRS), rely on trained examiners and can be time-consuming. Other assessments, such as brain imaging, are expensive and not always accessible. As a result, scientists have been searching for easier, more objective ways to measure disease progression.
Speech difficulties, known as dysarthria, are a common feature of Huntington’s and can appear before obvious motor symptoms. Therefore, a team of researchers in China investigated whether analyzing speech patterns with computer-based tools could serve as a reliable marker of disease severity.
The team recruited 141 people carrying a Huntington’s-causing mutation (carriers) and 69 healthy individuals without such mutations (controls). Among mutation carriers, 37 were not yet experiencing symptoms (premanifest disease), while the remaining 104 were (manifest Huntington’s).
All participants completed standard clinical and cognitive tests, and then read a standardized passage aloud into a smartphone. In addition, 96 carriers (both premanifest and manifest) underwent MRI scans to measure the volume of two brain structures, the caudate and putamen, which are strongly affected by Huntington’s and showed shrinkage early in the disease course.
Speech index score worsened from healthy controls to carriers
From the voice recordings, the team extracted 28 speech features, grouped into four categories: prosody (pitch and loudness), timing (speech rate and pauses), linguistic intelligibility (clarity and accuracy of speech), and voice quality (breathiness or roughness in the voice).
This four-feature speech model was able to distinguish carriers from healthy controls with an area under the curve, or AUC, of 0.901. An AUC of 1 represents a perfect discrimination between the two groups.
Instead of relying on individual speech features, the team developed a composite speech index using machine learning, a type of artificial intelligence that learns from data and detects patterns, allowing it to make predictions.
Here, the speech index combined the four most stable speech features: intensity (loudness) fluctuations, articulation and pronunciation accuracy, variability in pitch strength, and variability in speech segment durations (timing).
The speech index score rose (worsened) significantly from healthy controls to carriers, and across disease stages in a stepwise pattern, with significant increases from premanifest to manifest Huntington’s, and even higher increases between those with earlier and more advanced manifest disease.
Two of the four features, intensity fluctuations and articulation and pronunciation accuracy, increased as the disease advanced, while variability in pitch strength decreased.
We have developed and validated a machine learning-derived Speech Index as a sensitive, objective, and interpretable digital biomarker for HD severity.
The fourth feature, variability of speech segment durations, showed an unusual pattern. It decreased in premanifest and early-stage Huntington’s relative to healthy controls but rose again in more advanced stages, “suggesting a possible early compensatory mechanism that diminishes in later stages,” the researchers wrote.
Speech index scores were also significantly associated with scores on several standard measures of Huntington’s severity. These included the Total Motor Score, a part of the UHDRS that assesses motor impairment severity; the Symbol Digit Modalities Test, a part of the UHDRS that measures cognitive processing speed; and the composite UHDRS, which combines motor, cognitive, and functional abilities into one mathematical formula.
Higher speech index scores were also significantly associated with smaller caudate and putamen volumes.
These findings suggest that “speech characteristics captured by the Speech Index closely reflect both motor and cognitive impairment, as well as striatal [shrinkage], a core … feature of HD,” the team wrote.
Statistical models adjusted for age and sex confirmed these relationships, explaining between about 43% and 59.6% of the variability in these clinical and imaging outcomes.
Because the study was conducted at a single center in China with Mandarin-speaking participants, the team emphasized that broader testing in more diverse and multilingual groups is needed to confirm the tool’s general usefulness.
In addition, future studies will need to follow carriers over time to confirm that the speech index can track meaningful change as the disease progresses or in response to treatment.
“We have developed and validated a machine learning-derived Speech Index as a sensitive, objective, and interpretable digital biomarker for HD severity,” the researchers wrote, adding that its practicality “for remote administration” could make it useful for clinical trials and for monitoring patients between visits.
Leave a comment
Fill in the required fields to post. Your email address will not be published.