ANN-based Alzheimer's disease classification from bag of words

Alzheimer's disease (AD) is the most frequent cause of dementia and the patient numbers are increasing within an aging society. Prior research has shown that AD significantly affects the speech signal, and many approaches were published on how to detect AD from only speech or spoken text information. In an earlier work, we have proven the reliability of language models to statistically evaluate transcriptions from AD and healthy control participants. Based on these results, we propose the approach of counting word occurrences in transcriptions, storing them in a bag of words (BoW) vector, and using this vector as an input into an artificial neural network which classifies between AD and healthy state. It could be shown that the new method reached very similar results compared to the language model classifiers, although information about the word order was omitted.