Objective: The objective of this study is to establish a foundation for an automated differential diagnostic system for Parkinson’s disease (PD) and Parkinson-like disorders (PL) using motion capture technology.

Background: Distinguishing PD from PL is crucial for determining treatment strategies and predicting clinical prognosis. However, differential diagnosis remains challenging, even when combining detailed medical history with neurological findings. Advances in motion capture technology have enabled sophisticated motion analysis, which is expected to provide valuable support for the differential diagnosis.

Method: We enrolled 46 patients with PD (26 males, 20 females; median age, 62 years; interquartile range [IQR], 58-68) and 14 patients with PL (9 males, 5 females; median age, 67.5 years; IQR, 62-75; comprising 7 progressive supranuclear palsy, 5 multiple system atrophy, 2 idiopathic normal pressure hydrocephalus). During clinically defined medication off periods, participants performed a standardized battery of motor tasks, which included straight-line walking, turning, sit-to-stand and squat-to-sit maneuvers, as well as evaluations for postural, kinetic, and resting tremors.

Movements were recorded using a motion capture system. From the extracted coordinate data, we derived various kinematic features, including Euler angles of joints, angular velocities, and the volume of space occupied by the joints that moved. The dataset was partitioned into training and validation sets at a 2:1 ratio at the task level. A random forest method selected the top N features that maximized receiver operating characteristic curve (AUC) for clinical diagnosis prediction. These features were then used to build a LightGBM model. Model performance was evaluated by cross-validation AUC  (k = 5) on the training set and AUC on the validation set.

Results: In the training set, the model achieving mean AUC values of 0.86 for turning and 0.93 for straight-line walking. In the validation set, however, these values were 0.71 and 0.75, respectively. For all other tasks, the mean AUC remained below 0.70 in the validation phase. Six of the top ten features identified for straight-line walking demonstrated significant differences between the two groups.

Conclusion: Conclusions: Our findings demonstrate that gait analysis provides the highest diagnostic utility in the differential diagnosis of PD and PL.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/automated-differential-diagnosis-of-parkinsons-disease-and-parkinson-like-disorders-via-motion-capture-analysis/