Poster 133 - Quantification of Parkinson’s Motor Symptoms using an Optical-Based Contactless Leap Motion Controller: The Development of a Prototype System and an Algorithmic Approach
Student River Hill High School Clarksville, Maryland, United States
Introduction: Parkinson’s Disease (PD) is one of the fastest-growing neurological disorders worldwide. Currently, clinicians rely heavily on symptom assessment to fine-tune and evaluate the efficacy of treatments. Those assessments necessitate in-person visits, which can limit the frequency and accessibility. Traditional assessment tools, like the MDS-UPDRS-III, are often subjective and lack precision. Recent studies have explored wearable-based solutions, though these devices may not capture the fine-grained hand movements critical for accurate PD assessment. The overarching goal of this project is to design and develop a proof-of-concept application to collect data from patients participating in clinical studies. The core of this project is to design, develop, and evaluate a signal processing and machine learning-based algorithm to efficiently and accurately detect episodes of halting and hesitations during the “finger tapping” movement, a key component of the MDS-UPDRS-III assessment [1]. The underlying hypothesis is that the Leap Motion infrared sensor offers an affordable and precise method for assessing the severity of motor symptoms in PD. By leveraging the capabilities of the Leap Motion sensor, this project seeks to enhance the accuracy and accessibility of PD symptom evaluation, potentially transforming current assessment methodologies.
Materials and
Methods: Figure 1(a) illustrates the prototype system developed for collecting data for motor symptom assessment from PD patients in clinical settings. The system includes backend and frontend components integrated with the Leap Motion Controller. The backend system uses CMake, a powerful cross-platform linking tool, for managing the build process. It links and compiles the necessary code to produce executable binaries (.exe). We use UltraLeap SDK, which provides the essential libraries and tools for interfacing with the Leap Motion Controller. The front end is a graphical user interface (GUI) developed with PyQt. This interface allows clinicians to collect data and visualize and validate the collected data. Figure 1(b) is a segment of sample data collected with the system, annotated with hesitation episodes in orange.
To automatically detect the “hesitation” episodes, we developed an algorithm using an anomaly detection framework. The algorithm involves three steps: 1. The data is split into windows of 25 data points (~0.25 seconds) at a 100Hz sampling rate. For each window, a Homoscedasticity-Informed Correlation (H) is calculated, weighted by a kernel function (K), and summed to produce an anomaly score (A). N 2. Non-anomalous points are filtered out using the mean and standard deviation of A, 3. DBSCAN clustering is used to remove noise.
A grid search was conducted to fine-tune parameters alpha, beta, and gamma, which control the kernel function, statistical filtering, and DBSCAN, respectively.
Results, Conclusions, and Discussions: Figure 2 presents the model fitting results on N=15 sets of Leap Motion sensor time series (10s) collected from a healthy individual, annotated for hesitations and halts. We split the dataset into N=12 training sets for which optimized parameters were identified from grid search as illustrated in the heatmap on the left side. The model was then tested on the N=3 test set. The table on the right summarizes the training and test set results, comparing the proposed method and anomaly detection algorithm Isolation Forest. The proposed novel algorithm demonstrates promising results, particularly given the limited supply of training sets. It outperforms other popular anomaly detection algorithms, such as Isolation Forest.
This study demonstrates the potential value of the Leap Motion Controller-based system in enabling a real-world clinician to assess motor symptoms for PD patients precisely. The prototype system is fully functional and is currently being used to collect data in a real clinical setting in collaboration with a neurologist specializing in treating PD patients in Maryland.
Acknowledgements (Optional): I would like to acknowledge the mentoring support from Dr. Ramana Vinjamuri and Dr. Stephen Grill from Johns Hopkins University.
References (Optional): [1] C. G. Goetz et al., “Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS): Scale presentation and clinimetric testing results,” Movement Disorders, vol. 23, no. 15, pp. 2129–2170, 2008
[2] M. Ester, et al. "A density-based algorithm for discovering clusters in large spatial databases with noise.
