Gait classification in post-stroke patients using artificial neural networks

Gait Posture. 2009 Aug;30(2):207-10. doi: 10.1016/j.gaitpost.2009.04.010. Epub 2009 May 22.

Abstract

The aim of this study was to test three methods for classifying the gait patterns of post-stroke patients into homogenous groups. First, qualitative test results were found to correctly classify patients' gait patterns with an average success rate of 85%. Seeking further improvement, two quantitative methods were then tested. Analysis of min/max angle values in three lower limb joints, however, was less successful, showing a correct classification rate of below 50%. The best classification results were seen using an artificial neural network (ANN) to analyze the full progression of lower limb joint angle changes as a function of the gait cycle (with success rates from 100% for the knee joint to 86% for the frontal motion of the hip joint). These findings may help clinicians improve targeted therapy.

Publication types

  • Comparative Study
  • Evaluation Study
  • Research Support, Non-U.S. Gov't

MeSH terms

  • Adult
  • Aged
  • Arthrometry, Articular
  • Biomechanical Phenomena
  • Cluster Analysis
  • Diagnosis, Computer-Assisted*
  • Discriminant Analysis
  • Female
  • Gait Disorders, Neurologic / classification*
  • Gait Disorders, Neurologic / etiology
  • Gait Disorders, Neurologic / physiopathology*
  • Hemiplegia / etiology
  • Hemiplegia / physiopathology*
  • Humans
  • Male
  • Middle Aged
  • Neural Networks, Computer*
  • Stroke / complications*