A step forward in walking disability diagnosis

Background

Walking is a complex process that requires the precise coordination of muscles, joints, and the nervous system. When this coordination is disrupted - due to injury, neurological disorders, or other medical conditions - gait patterns can change significantly. Clinical gait analysis (CGA) is the established gold standard for assessing a person’s movement and identifying underlying movement disorders. Recent studies have emphasised the importance of CGA in medical decision-making, demonstrating its potential to improve patient outcomes by refining treatment strategies, increasing confidence and agreement in clinicians.

Clinical gait analysis uses motion capture technology, similar to what is used in the film and gaming industries. Small reflective markers are placed on key points of a patient’s body, such as the hip, knee, and ankle. Infrared cameras track these markers as the patient walks, and the recorded data is used to create a detailed, three-dimensional model of their movement. This allows clinicians to measure factors such as joint angles, step length, and walking speed with high precision. The results help doctors and therapists diagnose movement disorders, monitor disease progression, and evaluate the effectiveness of treatments. Despite its accuracy, clinical gait analysis is not error-free.

Some of the most common challenges are:

• Marker-placement error: Accurate analysis depends on precise marker positioning. Small variations in placement can lead to incorrect measurements.
• Soft-tissue artefacts: Because markers are placed on the skin, movements of soft tissue (such as muscle and fat) can cause slight shifts in marker position, leading to errors in data interpretation.
• Inaccurate identification of gait cycle events: a gait cycle consists of phases such as initial contact and foot off. Errors in detecting these phases can distort the analysis results, as they are used in the calculation of results (joint angles, step length, etc.) .

When errors exceed a certain threshold, the reliability of CGA data could be compromised and even lead to meaningless data.

Goals and Methods

The aim of this PhD project is to minimise the error source of inaccurate identification of gait cycle events using artificial intelligence. In addition, the knowledge gained will be made useable for everyday clinical practice with patients by creating an algorithm that will save time and reduce human error during post-processing of CGA data. With the aim of minimising the sources of error in 3DGA, in addition to reducing processing time, this project builds on experiences of other clinical applications of machine learning (e.g. image-based tumour identification).

First results

In the first publication, IntellEvent, a robust deep learning-based gait event detection for different pathologies, was created. The benefits of using IntellEvent lie in its high accuracy and robustness due to the large underlying dataset of more than 1000 patients from the Orthopaedic hospital Vienna-Speising. But why is a high accuracy important? Even small errors in gait event detection can introduce errors in the CGA results, which could change the interpretation.

Next steps

A multicenter study is planned, to evaluate how good IntellEvent performs on data from other laboratories. In addition to that, the IntellEvent-team is aiming to expand the algorithm usability to other movement tasks like turning, running, and stair climbing.

IntellEvent-team

Bernhard Dumphart, Djordje Slijepcevic, Andreas Kranzl, Fabian Unglaube, Matthias Zeppelzauer, Arnold Baca and Brian Horsak.

Publications

• The Effect of Inaccurate Initial Contact Events on Kinematics in Healthy and Pathological Gait
  B. Dumphart, D. Slijepcevic, F. Unglaube, A. Kranzl,  Baca, B. Horsak, (JDEc. 2024) In: Healthy and Pathological Gait. SSRN: ssrn.com/abstract=5095753 or dx.doi.org/10.2139/ssrn.5095753
• Robust deep learning-based gait event detection across various pathologies
  B. Dumphart, D. Slijepcevic, M. Zeppelzauer, A. Kranzl, F. Unglaube, A. Baca, B. Horsak, (Aug. 2023) In: PLOS, doi.org/10.1371/journal.pone.0288555.

Contact: bernhard.dumphart@univie.ac.at.