Objective: To investigate whether a virtual reality (VR)-based gait task can induce freezing of gait (FOG) in Parkinson’s disease (PD) and whether gait trajectory data can identify patients with FOG.

Background: Gait impairment in PD is strongly influenced by environmental factors, and FOG may emerge in confined spaces or doorways in daily life even when gait appears normal in clinical settings. This makes accurate clinical assessment challenging, and objective evaluation has not yet been established.

Method: Thirty patients with PD and 20 age-matched healthy controls performed a virtual hallway walking task using a head-mounted display. Five conditions were tested: four obstacle conditions combining passage width (narrow, 0.35 m; wide, 0.52 m) and obstacle arrangement (organized or cluttered), plus a no-obstacle baseline. Participants walked through a 3.0 × 1.0 m virtual hallway while avoiding box-shaped obstacles. Walking trajectories were obtained from headset-based head tracking, and movements were reconstructed as three-dimensional skeletal animations. FOG was rated on a 4-point scale from these videos. In the narrow cluttered condition, correlations were examined between FOG scores and the baseline-corrected number of hesitations. For Q-learning analysis, the walking space was divided into 25-cm grid cells, hesitation was treated as a penalty, cumulative penalty maps were generated, and receiver operating characteristic analysis was used to determine a threshold for identifying FOG.

Results: In the narrow cluttered condition, FOG scores correlated positively with the baseline-corrected number of hesitations (r = 0.677, p < 0.001). Q-learning analysis demonstrated concentrated penalties around obstacles, indicating spatial clustering of hesitation behavior. Receiver operating characteristic analysis showed strong discriminative performance, with an area under the curve of 0.932, sensitivity of 0.857, specificity of 0.891, and accuracy of 83.3%.

Conclusion: This VR-based gait assessment system induced FOG that is difficult to reproduce in routine clinical settings and enabled accurate quantification and identification of its spatial features using Q-learning analysis. These findings support its potential as an objective tool for FOG assessment in PD.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/automated-identification-of-freezing-of-gait-in-parkinsons-disease-using-q-learning-analysis-during-virtual-reality-based-gait-task/