Background: MCI is a heterogeneous condition with multiple subtypes. Although previous studies have investigated various speech and language characteristics of MCI-AD [cite], studies directly comparing both acoustic and lexical features across different MCI subtypes remain scarce, leaving an important gap in the literature. This study compares speech and language features in MCI due to Alzheimer’s disease (MCI-AD) to those in β-amyloid negative MCI with Lewy body (MCI-LB). 

 

Methods: Korean participants with MCI-AD (n=32, age=77.12±5.62 years, education=10.78±4.32 years) or MCI-LB (n=23, age=76.91±7.35 years, education=8.83±4.33 years) and healthy adults (HC; n=38, age=74.16±6.32 years, education=9.45±3.83 years) completed a speech collection protocol (Table 1). β-amyloid positivity was confirmed with visual assessment of 18F-Florbetaben PET. Recordings were automatically transcribed using ASR and manually corrected by trained annotators. We automatically extracted lexical features (e.g., counts of part-of-speech categories per 100 words) from a picture description task and acoustic features (e.g., loudness, pitch) from a sustained phonation task. Age and sex were covaried in all analyses, and p-values were adjusted using false discovery rate. 

 

Results: We found that individuals with MCI-LB exhibited breathier and less stable phonation, as reflected by lower harmonic-to-noise ratio values, and produced more interjections than both individuals with MCI-AD and HCs. In addition, the MCI-LB group differed significantly from HCs on several acoustic measures associated with hypophonia and reduced vocal fold control, whereas the MCI-AD group showed no significant differences from HCs on those measures. The MCI-LB group also differed significantly from HCs on a number of lexical features, and the MCI-AD group also showed a significant lexical difference from HC in the count of auxiliaries. Furthermore, several of the acoustic and lexical features were significantly associated with clinical measures of cognitive and motor functioning. 

 

Conclusion: Individuals with MCI-LB exhibited breathier and less stable phonation, as reflected by lower harmonic-to-noise ratio values, and produced more interjections than both individuals with MCI-AD and HCs. The MCI-LB group also differed significantly from HCs on several acoustic measures associated with hypophonia and reduced vocal fold control, whereas the MCI-AD group showed no significant differences from HCs on those measures. The MCI-LB and MCI-AD groups also differed significantly from HCs on a number of lexical features. Several of the acoustic and lexical features were associated with clinical measures of cognitive and motor functioning. Results highlight the potential of using speech as an early screening tool of patients with different MCI subtypes