Dysarthria Subgroups in Talkers with Huntington's Disease: Comparison of Two Data-Driven Classification Approaches

Although researchers have recognized the need to better account for the heterogeneous perceptual speech characteristics among talkers with the same disease, guidance on how to best establish such dysarthria subgroups is currently lacking. Therefore, we compared subgroup decisions of two data-driven approaches based on a cohort of talkers with Huntington's disease (HD): (1) a statistical clustering approach (STATCLUSTER) based on perceptual speech characteristic profiles and (2) an auditory free classification approach (FREECLASS) based on listeners' similarity judgments. We determined the amount of overlap across the two subgrouping decisions and the perceptual speech characteristics driving the subgrouping decisions of each approach. The same speech samples produced by 48 talkers with HD were used for both grouping approaches. The STATCLUSTER approach had been conducted previously. The FREECLASS approach was conducted in the present study. Both approaches yielded four dysarthria subgroups, which overlapped between 50% to 78%. In both grouping approaches, overall bizarreness and speech rate characteristics accounted for the grouping decisions. In addition, voice abnormalities contributed to the grouping decisions in the FREECLASS approach. These findings suggest that apart from overall bizarreness ratings, indexing dysarthria severity, speech rate and voice characteristics may be important features to establish dysarthria subgroups in HD.

Automated summary

Comparing two data-driven approaches for subgrouping HD patients, researchers found 50% to 78% overlap between the subgroup decisions. Overall strangeness and speech rate are important factors for grouping decisions in both approaches. Voice abnormalities also contribute to grouping decisions in one approach.