Article Text
Abstract
Introduction Fluency tests are widely employed in the assessment of cognition. We assessed the feasibility of integrating these tasks into an automated cognitive assessment tool; the ‘Digital Doctor’.
Methods 15 each of healthy controls (HC), Alzheimer’s disease (AD) and mild cognitive impairment (MCI) were recruited. Participants named as many words beginning with P (phonemic fluency) and as many animals (semantic fluency) as they could within one minute.
Results Manual Analysis: HC named the same number of words for both tasks compared to normative data. Patients with AD and MCI were able to complete the task and Z-scores showed significant decreases in category fluency between the AD and HC groups, -2.25.
Automated Analysis: Automatic Speech recognition (ASR) was 77.9% accurate for semantic testing and 64.8% for phonemic testing, ASR overestimated the mean scores in the AD and MCI groups. Comparison of phonemic fluency in AD compared to HC yielded a significant Z score, -2.17.
Conclusions Automated collection of fluency data is feasible and comparable to manually collected data. The over-estimation of the number of correct responses provided by the ASR will be improved with more data along with training of the Machine Learning algorithms to recognise certain, expected words.