Adonay Nunes1, Ram Kinker Mishra1, Jose Casado1, Gozde Cay2, Mohammad Dehghan Rouzi2, Myeounggon Lee2, Ashkan Vaziri1, Bijan Najafi2
1BioSensics LLC, 2Baylor College of Medicine
Objective:
To demonstrate the efficacy of digital speech analysis as a tool for identifying and gauging cognitive impairment in older adults by focusing on speech biomarkers.
Background:
Current cognitive assessments face challenges such as floor/ceiling and practice effects, insufficient psychometric performance in milder cases, and repeated assessment effects. Digital speech analysis presents an opportunity to overcome these issues, especially in the field of detecting cognitive decline.
Design/Methods:
Older adults with varied cognitive states were recruited. Speech data was recorded as participants read a standard passage aloud, and then processed with BioDigit Speech to derive digital biomarkers, including timing, pitch, and intelligibility. Group differences were measured using Cohen's d, and correlations were established with the Montreal Cognitive Assessment (MoCA). A stepwise approach employing a Random Forest model was utilized to classify cognitive decline vs cognitive intact individuals through speech features.
Results:
The study involved 59 participants: 36 with cognitive impairment and 23 cognitively intact controls. Among all parameters, similarity determined by Dynamic Time Warping (DTW) displayed the strongest positive correlation (rho=0.529, p<0.001) with MoCA scores, while timing parameters, specifically the ratio of extra words, exhibited the most potent negative correlation (rho=-0.441, p<0.001). Discriminative performance peaked using a combination of four speech parameters: total pause time, speech to pause ratio, similarity DTW, and the ratio of extra words. The precision and balanced accuracy scores recorded were 84.3 ± 1.5% and 75.0 ± 1.4%, respectively.
Conclusions:
Speech data analysis effectively distinguishes between cognitively impaired and cognitively intact older adults. Such digital speech analysis holds promise as a biomarker for early detection and continual monitoring of cognitive decline, paving the way for innovative strategies in dementia care.