Facial Feature Analysis for Pseudodementia: A Preliminary Study

Brian Sumali¹, Yasue Mitsukura¹, and Taishiro Kishimoto²

1. Department of System & Design Engineering, Faculty of Science and Technology, Keio University, Japan

2. Department of Neuropsychiatry, School of Medicine, Keio University, Japan

Abstract: Pseudodementia is a type of temporary cognitive impairment caused by mental health disorders, differing from true dementia with root cause of neurological disorders. The most common cause is depression, and the comorbidity of dementia with depression in elderly patients deludes even expert psychologists. Although pseudodementia can be diagnosed with extensive testing, it is time consuming and taxing for both the psychiatrists and the patient, as at least two tests must be taken — one for depression screening and another for dementia screening. Additionally, although machine learning has been utilized in automated mental health screening, attention for pseudodementia is minimal, with progression only in discussing and proposing pseudodementia tests in the medical field. In this research we aim to extract facial features from actual dementia patients and depression patients, which were conducted and diagnosed by licensed clinical psychiatrists. The features extracted from dementia patients and depression patients served as the basis for screening pseudodementia. We also tried to utilize machine learning using the facial features to examine the possibility of automated pseudodementia screening. As it is a preliminary study, a conventional machine learning model was utilized along with popular feature selection algorithm. Satisfactory result of 81.7% accuracy was obtained, although improvement must be made for actual clinical implementation. Several features that were chosen by the machine learning was also reported, which may be beneficial for human clinician.

Key words: pseudodementia, facial features, machine learning