Automatic detection of mild cognitive impairment and dementia through facial expression analysis

This thesis presents an automatic system for cognitive impairment detection from facial expressions in people with mild cognitive impairment and dementia. Dementia is one of the main causes of disability and dependency in the elderly worldwide, leading to impairment in cognitive functions, and its global prevalence is a pressing concern for national health-care systems. Currently, there are no approved disease-modifying therapies. Early detection is thus crucial for the well-being of patients and their caregivers, and to enable patients’ inclusion in research and clinical trials for therapies in development. A precursor stage of dementia known as mild cognitive impairment (MCI) has been identified, in which daily activities are still not greatly affected. However, in today’s context achieving both high accuracy and low cost in the early detection of cognitive impairment poses a difficult task. Multiple studies have provided evidence that cognitively impaired individuals exhibit abnormal facial expressions when elicited to manifest target emotions. Therefore, our proposed AI-based system analyzes the evolution of emotional states inferred from facial expressions to differentiate between cognitively impaired and healthy individuals. Through training on data collected by our research group, which included both MCI and dementia patients besides healthy controls, the system successfully achieved the classification accuracy of 74.4 ± 5.15%. These results suggest the feasibility of utilizing facial expressions as potential early markers for cognitive impairment and lay the foundation for further research in this field.