Statistical Analysis of Dual-task Gait Characteristics for Cognitive Score Estimation

Taku Matsuura, Kazuhiro Sakashita, Andrey Grushnikov, Fumio Okura, Ikuhisa Mitsugami, Yasushi YagiScientific 2019 · DOI

TL;DR

This paper introduces an automated system using Kinect to analyze gait and cognitive task performance during single and dual-task scenarios. The system successfully estimates cognitive scores (MMSE) by identifying key gait characteristics, offering a more frequent and less intrusive screening method for cognitive decline compared to traditional paper tests.

What Problem Does This Paper Address?

Traditional methods for screening cognitive impairment, like the Mini-Mental State Examination (MMSE), rely on paper-based tests. While effective, these tests are time-consuming, require a human administrator, and are not suitable for frequent use, potentially delaying the detection of early cognitive decline. Dementia significantly impacts multitasking abilities, suggesting that dual-task performance could be a sensitive indicator of cognitive impairment. However, previous research in this area has often used limited datasets and simple gait features. This paper aims to bridge this gap by developing and validating an automated system that leverages dual-task gait analysis for more efficient and frequent cognitive score estimation.

Methods

The core of this research is an automated dual-task acquisition system built around a Microsoft Kinect v2 sensor. This system guides participants through three phases: a single cognitive task (calculus), a single physical task (marching in place), and a dual-task combining both. During these tasks, the Kinect captures body skeleton data, and other sensors record response times and accuracy for the cognitive task.

From this data, 12 features are extracted, encompassing gait characteristics (e.g., walking speed, knee angle, and their standard deviations) and cognitive performance metrics (e.g., correct answer rate, answer speed). These features are then fed into machine learning algorithms (Support Vector Machine, Random Forest, and Neural Networks) for MMSE score estimation. The primary goal was to classify subjects into two groups: those with an MMSE score below 24 (indicating potential cognitive impairment) and those with 24 or above. The researchers employed a leave-one-subject-out cross-validation strategy for robust evaluation.

Datasets

The study collected data from 90 subjects across three elderly facilities. Over a year of operation, the system gathered 10,833 behavior data samples. The ground truth for cognitive function was established using the MMSE, administered face-to-face. The paper does not introduce a new benchmark dataset but rather utilizes real-world data collected from elderly individuals.

Key Insights and Ideas

  • Dual-tasking as a sensitive cognitive indicator: The research strongly supports the idea that the breakdown in performance during dual-tasking, particularly in gait and cognitive coordination, is a significant marker for cognitive decline.
  • Gait characteristics are crucial: Beyond just walking speed, subtle gait features like the standard deviation of walking speed during dual-tasking and knee angle (both single and dual-task) emerged as highly important predictors of MMSE scores.
  • Automated systems offer advantages: The proposed Kinect-based system provides a fully automatic, operator-free solution that is significantly faster (90 seconds) than traditional paper tests, enabling more frequent monitoring.
  • Learning effect in gait: Participants with higher MMSE scores showed a tendency to increase their knee lift after multiple trials. This suggests that while cognitive impairment might hinder this adaptation, the physical act of consciously adjusting gait might be linked to cognitive awareness.

Key Results

  • The Neural Network (NN) model achieved the best performance for MMSE score estimation and classification, yielding a combined recall and specificity of 1.552.
  • The NN model successfully classified subjects with an MMSE score of 23 or lower with a recall of 0.753 and a specificity of 0.799.
  • The study identified the standard deviation of walking speed during dual-task and knee angle (single and dual-task) as the most important features for MMSE score estimation.
  • Combining single-task and dual-task features led to improved classification accuracy compared to using either alone.
  • The accuracy of classification increased with the number of trials conducted by the subjects, suggesting a learning or adaptation effect.

Limitations and Future Work

  • Limited cognitive task variety: The current cognitive task involves simple calculus problems. The authors suggest exploring more complex or different types of cognitive assignments to potentially detect milder forms of cognitive impairment like Mild Cognitive Impairment (MCI).
  • Generalizability: While the study used data from three facilities, further validation across diverse populations and settings would be beneficial.
  • Subjective awareness: The observed increase in knee lift with experience raises questions about whether this is purely a learned motor adaptation or reflects a heightened cognitive awareness of physical performance, which warrants further investigation.
  • Integration with other metrics: Future work could explore combining dual-task gait analysis with other cognitive assessment tools (e.g., MoCA) or medical information (e.g., fMRI) for a more comprehensive diagnostic approach.

Summary updated

Abstract

Traditional approaches for the screening of cognitive function are often based on paper tests, such as Mini-Mental State Examination (MMSE), that evaluate the degree of cognitive impairment and provide a score of patient's mental ability. Procedures for conducting paper tests require time investment involving a questioner and not suitable to be carried out frequently. Previous studies showed that dementia impaired patients are not capable of multi-tasking efficiently. Based on this observation an automated system utilizing Kinect device for collecting primarily patient's gait data who carry out locomotion and calculus tasks individually (i.e., single-tasks) and then simultaneously (i.e., dual-task) was introduced. We installed this system in three elderly facilities and collected 10,833 behavior data from 90 subjects. We conducted analyses of the acquired information extracting 12 features of single- and dual-task performance developed a method for automatic dementia score estimation to investigate determined which characteristics are the most important. In result, a machine learning algorithm using single and dual-task performance classified subjects with an MMSE score of 23 or lower with a recall 0.753 and a specificity 0.799. We found the gait characteristics were important features in the score estimation, and referring to both single and dual-task features was effective.
BibTeX
@inproceedings{matsuura2019statistical,
  author = {Taku Matsuura and Kazuhiro Sakashita and Andrey Grushnikov and Fumio Okura and Ikuhisa Mitsugami and Yasushi Yagi},
  title = {{Statistical Analysis of Dual-task Gait Characteristics for Cognitive Score Estimation}},
  booktitle = {Scientific},
  year = {2019},
  doi = {10.1038/s41598-019-56485-w},
}
Figure