Machine vision-based gait scan method for identifying cognitive impairment in older adults

Area of Science:

• Gerontology
• Computer Vision
• Neurology

Background:

• Early identification of cognitive impairment is crucial for managing age-related disabilities.
• Gait parameters are recognized as indicators and predictors of cognitive decline.
• Existing machine learning methods for gait analysis in cognitive studies require optimization, particularly using machine vision.

Purpose of the Study:

• To develop and evaluate a novel deep machine vision-based method for analyzing gait to identify cognitive decline in older adults.
• To introduce the Deep Optimized GaitPart (DO-GaitPart) network for enhanced gait pattern recognition.

Main Methods:

• Utilized the West China Hospital Elderly Gait dataset (158 adults) labeled with cognitive status (Short Portable Mental Status Questionnaire).
• Proposed the DO-GaitPart network employing silhouette and skeleton gait images.
• Incorporated a short-term temporal template generator (STTG), depth-wise spatial feature extractor (DSFE), and multi-scale temporal aggregation (MTA) with an attention mechanism.

Main Results:

• Ablation tests confirmed the essential contribution of each DO-GaitPart component.
• DO-GaitPart demonstrated superior performance on the CASIA-B and Gait3D datasets compared to existing methods.
• Achieved a Receiver Operating Characteristic Area Under the Curve (ROCAUC) of 0.876 for cognitive state classification.

Conclusions:

• The DO-GaitPart method effectively identifies cognitive decline from gait video data.
• This machine vision approach serves as a potential prototype tool for cognitive assessment in older adults.