Speech-Based Dementia Classification for FTLD Diagnosis Support

This paper proposes a screening system to automatically detect a Frontotemporal Lobar Degeneration (FTLD) and support a diagnosis of a general practitioner. Dementia results from a variety of diseases that primarily or secondarily affect the brain. It is important to diagnose an underlying disease correctly. We have been investigating FTLD, which is one of diseases. We took into account the specific symptoms, used speech features to classify FTLD, Alzheimer's disease (AD) and healthy control (HC). We confirmed that our method can classify three groups with accuracy of 0.84 and macro F-measure of 0.79. We also showed the effectiveness of linguistic features in FTLD detection.