Abstract

The COVID-19 pandemic has abruptly and undoubtedly changed the world we knew at the end of the second decade of the 21st century. To be prepared for future possible pandemics, special attention should be devoted to the fact that seniors, aged 60 and over, are more vulnerable during the pandemic era. In addition to a higher risk of infections, seniors are also at higher risk of suffering from mental and cognitive issues. Given an expected and alarming population ageing in near future, it is crucial and of significant importance to developing innovative and advanced autonomous cognitive screening systems. Of particular interest to this thesis is the development of autonomous cognitive screening systems via the integration of Signal Processing (SP), Artificial Intelligence (AI), and Machine Learning (ML) models. In particular, the focus is on the development of an AI-empowered avatar that autonomously performs the Neurobehavioral Cognitive Status Examination (Cognistat) assessments. Results obtained from Cognitive screening tests can be used in conjunction with other data sources to perform differential diagnosis of dementia or other cognitive disorders. Although Cognistat is widely utilized in clinical applications, its administration and interpretation of the results solely rely on a well trained physiologist, which is highly restrictive during a pandemic. Towards addressing this issue, in the thesis, an Automated Virtual Cognitive Assessment (AVCA) framework [1, 2] is proposed that integrates Natural Language Processing (NLP) and hand gesture recognition techniques. The AVCA framework is an autonomous cognitive assessment system that receives audio and video signals in a real-time fashion and performs semantic and synthetic analysis using NLP techniques and DNN models. The proposed framework provides individual scores in the seven major cognitive domains, i.e., orientation, attention, language, contractual ability, memory, calculation, and reasoning. Additionally, we propose an efficient model to facilitate human-machine interactions from speech recognition to text classification. In particular, an unsupervised contrastive learning framework is proposed using Bidirectional Encoder Representations from Transformers (BERT) that outperforms its state-of-the-art unsupervised counterparts.