Semantic verbal fluency (SVF) tests are routinely used in screening for mild cognitive impairment (MCI) and seem to be highly sensitive to cognitive changes [1]. In this task, participants name as many items of a semantic category under a time constraint. SVF can be considered as a multifactorial task, comprising both semantic memory retrieval and executive control processes. However, clinicians measure task performance manually only by summing the number of correct words and errors. More fine-grained variables add valuable information to clinical assessment, but are time-consuming. Therefore, the aim of this study is to investigate whether automatic speech analysis of the SVF could provide these as accurate as manual and thus, support qualitative screening of neurocognitive impairment. Furthermore, we examined the technologic feasibility of automatically assessing the SVF task – via a telephone-based solution for potential remote frontline prescreening of cognitive impairments [2]. We will present SVF data which was collected from 95 older people with MCI (n=47), Alzheimer’s or related dementias (ADRD; n=24) and healthy controls (HC; n=24). Speech was recorded through a mobile tablet device using the built-in microphone. All data was annotated manually and automatically with the named words, clusters and switches. The obtained metrics were validated using a classifier to distinguish HC, MCI and ADRD. Afterwards to simulate telephone conditions, the recordings were downsampled to a lower sampling rate. Automatically extracted clusters and switches were highly correlated (r=0.9) with manually established values, and performed as well on the classification task separating healthy controls from persons with Alzheimer’s (AUC=0.939) and MCI (AUC=0.758) [3]. The downsampled quality obtained similar encouraging results. We observe a relatively low word error rate of 33% despite phone-quality speech samples. The automated classification pipeline performs equally well compared to the classifier trained on manual transcriptions of the same speech data. Our results indicate SVF as a prime candidate for inclusion into an automated telephone-screening system and that it is possible to automate fine-grained analyses of this task for the assessment of cognitive decline.
