Y9 - Uncovering Wrist Posture Effects on Finger Movement Decoding: Implications for Prosthetic Control

Saturday, October 11, 2025

10:00 AM - 11:00 AM PDT

Location: Exhibit Hall F, G & H

Co-Author(s)

SP

Sreya Pothuraju

Graduate Student Researcher
University of Pittsburgh, United States

Primary Investigator(s)

RG

Robert Gaunt

Associate Professor
University of Pittsburgh
Pittsburgh, Pennsylvania, United States

Presenting Author(s)

AC

Anabelle Creveling

Student
University of Rochester, United States

Introduction: : Over one-third of upper-limb prostheses are rejected by people living with an amputation, in part due to limited functionality [1], including insufficient finger and wrist control [2]. Current commercial prosthetic hands offer restricted wrist motion and lack support for natural, independent control of individual fingers. A key challenge in implementing this more complex digit and wrist control is understanding how different wrist postures impact electromyography (EMG) based decoders. EMG activity from the same finger movements can vary by up to 70% across wrist postures [3], introducing variability that complicates EMG interpretation. This presents a significant obstacle, as decoders are often trained on EMG data collected under a limited range of wrist postures, and in many cases, postural effects are not explicitly accounted for.
Here, we systematically evaluate how wrist posture affects finger movement decoding. We used linear and nonlinear dimensionality reduction techniques to study intramuscular EMG data across multiple wrist postures to reveal the underlying structure and separability of the data. To determine whether the changes in finger muscle activity across wrist postures are driven by wrist muscle contributions or reflect inherent modulation of finger muscles themselves, we performed the analysis on two datasets: one including all forearm muscles and another including only the extrinsic finger muscles. These findings will inform the development of more robust myoelectric decoders for upper limb prostheses.

Materials and

Methods: : Data from a previous study were used, comprising intramuscular EMG recordings from forearm muscles in 10 able-bodied individuals [3]. Participants performed 10 repetitions of isolated finger flexion-extension movements across five different wrist postures: neutral, flexed, extended, pronated, and supinated. Our goal was to examine how the underlying structure of EMG representations changes when performing the same task under different postures. To study this, we used Principal Component Analysis (PCA), a linear dimensionality reduction method, as well as Uniform Manifold Approximation and Projection (UMAP), a non-linear dimensionality reduction method. We assessed the separability of wrist posture manifolds across a different number of dimensions by using an unsupervised clustering approach, Gaussian Mixture Model (GMM) clustering, and quantified clustering quality using silhouette scores. These scores reflect how well a data point fits within its assigned cluster relative to neighboring clusters, ranging from 1 (perfect separation) to −1 (misclassification). To evaluate posture-specific clustering, GMM-derived labels were compared to the true wrist posture labels, allowing us to identify which postures were confused in the reduced-dimensional space. To assess how these differences in structure impact decodability, we trained two types of decoders: a linear decoder, Linear Discriminant Analysis (LDA), and a nonlinear decoder, Support Vector Machine (SVM) with a radial basis function kernel, to predict finger identity from EMG data. We used a 70:30 train-test split with 100-fold Monte Carlo cross-validation across all participants.

Results, Conclusions, and Discussions:: We compared the cluster labels obtained from GMM clustering to the true movement labels and found that the clusters in low dimension PCA space did not align well with the true labels (Figure 1a). In contrast, the clusters derived from the UMAP embedding were closely aligned with the true labels (Figure 1c). This suggests that UMAP better preserves the posture-relevant structure in EMG data, allowing for a more accurate unsupervised classification through GMM clustering. We then examined the latent trajectories over time for single digit movement across all postures; in principle component space, these movements are represented by multiple clusters (Figure 1b), while the UMAP dimensions are constrained to a single unique cluster (Figure 1d). Silhouette scores were consistently higher when using UMAP compared to PCA, indicating improved separability of wrist posture manifolds across all 16 dimensions (Figure 2). This suggests superior clustering performance where each posture can be uniquely identified using UMAP.
To determine whether the changes in finger muscle activity across wrist postures were driven by wrist muscle contributions or reflect inherent modulation of finger muscles themselves, we performed the analysis on two datasets: one including all forearm muscles and another including only the extrinsic finger muscles. Similar trends in low-dimensional structure and silhouette scores were observed even after excluding wrist muscles from the dataset, indicating that extrinsic finger muscle activity itself varies systematically with wrist posture. The decoder analysis also showed that the nonlinear decoder, SVM, significantly outperformed the linear decoder, LDA, on the combined data set, highlighting the influence of wrist posture on EMG-based decoding (Figure 3). Together, these findings indicate that wrist posture introduces nonlinear variability in finger EMG signals, and that accounting for this by combining multi-postural data and applying nonlinear methods such as UMAP can significantly enhance decoding performance for prosthetic applications.

Acknowledgements and/or References (Optional): :

References:
[1] Biddiss, E., & Chau, T. (2007). Upper-limb prosthetics: critical factors in device abandonment. American journal of physical medicine & rehabilitation, 86(12), 977–987. https://doi.org/10.1097/PHM.0b013e3181587f6c
[2] Rekant, J., Fisher, L. E., Boninger, M. L., Gaunt, R. A., & Collinger, J. L. (2023). Amputee, clinician, and regulator perspectives on current and prospective upper extremity prosthetic technologies. Assistive Technology, 35(3), 258-270.
[3] Beringer, C.R., Mansouri, M., Fisher, L.E. et al. The effect of wrist posture on extrinsic finger muscle activity during single joint movements. Sci Rep 10, 8377 (2020). https://doi.org/10.1038/s41598-020-65167-x