Detecting Dementia from Patients’ Conversational Transcripts: A Neural Deep Network Approach

Neurogenerative disorders such as Alzheimer’s Disease (AD) and related dementias are a growing problem since the global population is aging more and more. It is important to develop automated methods that can be aid in identifying the first symptoms of these diseases to better treat them from their earliest stages. Speech and language alterations are one of the first signs of dementia and this work is focused on developing an automated methodology that starting from patients’ linguistic samples can spot the presence of linguistic patterns that are related to dementia of the AD type. We are comparing different neural network models that starting from patients’ conversational transcripts use syntactic and semantic features to classify AD patients and Healty Control (HC) patients. We are then performing feature selection to understand what kind of feature plays a more significant role in the classification. Finally, we are interpreting some portions of the models proposed to analyze specific linguistic patterns linked with dementia of the AD type.