Use of electroencephalogram, gait, and their combined signals for classifying cognitive impairment and normal cognition

Abstract

BackgroundEarly identification of people at risk for cognitive decline is an important step in delaying the occurrence of cognitive impairment. This study investigated whether multimodal signals assessed using electroencephalogram (EEG) and gait kinematic parameters could be used to identify individuals at risk of cognitive impairment.MethodsThe survey was conducted at the Veterans Medical Research Institute in the Veterans Health Service Medical Center. A total of 220 individuals volunteered for this study and provided informed consent at enrollment. A cap-type wireless EEG device was used for EEG recording, with a linked-ear references based on a standard international 10/20 system. Three-dimensional motion capture equipment was used to collect kinematic gait parameters. Mild cognitive impairment (MCI) was evaluated by Seoul Neuropsychological Screening Battery-Core (SNSB-C).ResultsThe mean age of the study participants was 73.5 years, and 54.7% were male. We found that specific EEG and gait parameters were significantly associated with cognitive status. Individuals with decreases in high-frequency EEG activity in high beta (25–30 Hz) and gamma (30–40 Hz) bands increased the odds ratio of MCI. There was an association between the pelvic obliquity angle and cognitive status, assessed by MCI or SNSB-C scores. Results from the ROC analysis revealed that multimodal signals combining high beta or gamma and pelvic obliquity improved the ability to discriminate MCI individuals from normal controls.ConclusionThese findings support prior work on the association between cognitive status and EEG or gait, and offer new insights into the applicability of multimodal signals to distinguish cognitive impairment.

The Associations between Cognitive Function, Depressive Symptoms, and Contact with Adult Children in Older Couples

“…A higher percentile score indicates better cognition. Participants scoring ≤ 16th percentile were considered cognitively impaired [ 23 , 24 ].…”

Multimodal learning of clinically accessible tests to aid diagnosis of neurodegenerative disorders: a scoping review

“…For classifying healthy controls from those with AD, multimodal studies using MRI and PET data reported accuracy ranging from 74.3 [ 33 ] to 98% [ 52 ]. For discriminating MCI, AD and HC, the accuracy was generally lower with accuracies ranging from about 72–86% [ 52 , 55 ]. For classifying PD and HC with combined MRI and PET scan data, accuracies ranged from 88.57 [ 44 ] to 98.17% [ 31 ].…”

“…There are 4 studies related to AD and 7 studies related to PD. The most common combination of data was movement data and speech (n = 5 studies) with one study focusing on AD [ 47 ] and 4 on PD [ 29 – 31 , 34 ], followed by movement data and EEG signals (n = 3 studies, with 1 study in AD and 2 in PD) [ 26 , 55 , 66 ].…”

“…Two AD related studies [ 47 , 55 ] and 4 PD [ 29 , 30 , 32 , 66 ] studies used gait (walking) movement data. Generally, gait data was collected whilst participants walked at their usual pace over nine meters with a marker-based motion capture system (camera or sensor-based).…”

“…For classification of AD and MCI from HC, the gait data was often combined with speech and drawing modalities with accuracies ranging from 0.73 to 0.93 [ 47 , 55 ]. Yamada [ 47 ] focused on AD and MCI and used gait, speech, and drawing behaviours to classify patients from MCI and HC.…”