Using Loss of Balance to Understand Postural Instability Associated With Parkinson's Disease

Postural instability compromises balance, contributing to annual fall rates of 45%-68% in Parkinson's disease (PD). A fundamental gap in clinical evaluation and treatment of postural instability is the use of insufficiently challenging postural control tests and reliance on subjective scoring. Traditional 3D motion capture (Traditional-MC), while precise and objective, is not feasible for clinical utilization. Markerless motion capture (MMC) is a viable candidate for quantifying postural control, feasible with embedded cameras of augmented reality (AR) headsets. This project aimed to assess MMC accuracy in quantifying postural control and to compare PD patients and healthy controls (HCs) to develop a loss-of-balance (LOB) prediction model.

Video data were acquired by a clinician wearing a Microsoft HoloLens 2 AR headset as participants completed three progressively challenging postural control stances. Depth and red-green-blue camera data were analyzed with our custom-built human pose estimation software (CART-MMC). Criterion validity between Traditional-MC and CART-MMC was completed on outcomes, 95% ellipsoid sway area (Sway area) and 95% mediolateral and anteroposterior (ML and AP) ranges in 54 (HC = 28, PD = 26) participants. Group differences were assessed, and survival analysis was conducted to quantify no LOB probabilities in 31 HCs and 68 PD patients. A relaxed LASSO model was used to predict LOB in the PD group from CART-MMC data.

Sway area and ML range from CART-MMC were equivalent to Traditional-MC. Furthermore, ML range from CART-MMC differentiated PD patients from HCs in the stance that had the highest completion rate. As postural task difficulty increased, fewer PD patients were able to complete compared to HCs, resulting in lower survival probabilities of the PD group in Tandem-Firm-EO and FT-Foam-EC stances. In the least challenging stance, ML range predicted the occurrence of LOB in the more challenging stance (LOO-CV AUC = 0.78).

CART-MMC is a valid method of obtaining objective postural control outcomes, which can detect latent postural control deficits and demonstrates a scalable, objective clinical and remote monitoring approach for predicting falls.