Category: Parkinson's Disease: Cognitive functions
Objective: To identify speech acoustic measures associated with mild cognitive impairment in PD (PD-MCI).
Background: Detection of PD-MCI is a challenge for clinical care and research. There are no easy-to-use, scalable approaches validated to screen for and monitor PD-MCI. However, speech acoustic analysis has shown promise in monitoring other aspects of PD, including prodromal and motor symptoms. In this work, we identified speech acoustic markers of PD-MCI using a picture description task.
Method: In a cross-sectional study, participants with PD-MCI (n=24) and normal cognition (PD-NC, n=18) were audio-recorded while describing the Cookie theft picture and reading a passage. PD-MCI was diagnosed by MDS Task Force Tier II Criteria. MDS-UPDRS Part III was categorized into mild (UPDRS <= 32, n=25) and moderate symptoms (32 < UPDRS < 59, n=17). Speech raw features were formants, Mel-frequency cepstral coefficients (MFCCs), cepstral peak prominence (CPP), and envelope (Env) as well as their first-order derivatives (dFormants, dMFCC, dCPP, dEnv). Statistical features were computed from each raw feature channel. Eigenspectral features were computed from multichannel correlation matrices constructed with time delay embedding. Features were input into a Gaussian mixture model classifier to detect either PD-MCI or moderate motor symptoms. Detection accuracy with cross-validation was quantified using the area under the ROC curve statistic (AUC).
Results: The picture task yielded acoustic measures with high accuracy in discriminating PD-MCI from PD-NC. Several features yielded AUC >0.7 including five of the raw features (MFCC, dMFCC, Env, dEnv, CPP) and 2 eigenspectral features (MFCC, dMFCC). For the reading task, only Env and dCPP yielded AUC > 0.7 for discrimination of cognitive status. Optimal discrimination model performance was AUC=0.86 using all picture task features with AUC>0.7. Acoustic features were only modestly associated with motor severity. One feature (dFormants) yielded AUC >0.7 for discriminating mild vs. moderate symptoms, which occurred for both tasks.
Conclusion: This work adds to the emerging potential of speech acoustic analysis as a powerful tool to detect and monitor various symptoms in PD. Our approach has high accuracy in discriminating PD-MCI from PD-NC compared to rigorous criteria, and may be clinically useful as a scalable and low burden screening tool.
To cite this abstract in AMA style:
K. Smith, J. Williamson, T. Quatieri. Detecting Mild Cognitive Impairment in Parkinson’s Disease using Speech Markers [abstract]. Mov Disord. 2024; 39 (suppl 1). https://www.mdsabstracts.org/abstract/detecting-mild-cognitive-impairment-in-parkinsons-disease-using-speech-markers/. Accessed October 7, 2024.« Back to 2024 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/detecting-mild-cognitive-impairment-in-parkinsons-disease-using-speech-markers/