Linguistic-Based Mild Cognitive Impairment Detection Using Informative Loss

This paper presents a deep learning method using Natural Language Processing(NLP) techniques, to distinguish between Mild Cognitive Impairment (MCI) andNormal Cognitive (NC) conditions in older adults. We propose a framework thatanalyzes transcripts generated from video interviews collected within theI-CONECT study project, a randomized controlled trial aimed at improvingcognitive functions through video chats. Our proposed NLP framework consists oftwo Transformer-based modules, namely Sentence Embedding (SE) and SentenceCross Attention (SCA). First, the SE module captures contextual relationshipsbetween words within each sentence. Subsequently, the SCA module extractstemporal features from a sequence of sentences. This feature is then used by aMulti-Layer Perceptron (MLP) for the classification of subjects into MCI or NC.To build a robust model, we propose a novel loss function, called InfoLoss,that considers the reduction in entropy by observing each sequence of sentencesto ultimately enhance the classification accuracy. The results of ourcomprehensive model evaluation using the I-CONECT dataset show that ourframework can distinguish between MCI and NC with an average area under thecurve of 84.75