Bulea – Patwardhan – 2026

Machine Learning Models for Predicting Human Movement from Ultrasound Imaging

The Neurorobotics Research Group is developing novel machine learning-based approaches to estimate human movement from real-time ultrasound imaging. This project builds on a rich, previously collected multimodal dataset consisting of time-synchronized ultrasound images, motion capture, electromyography (EMG), and exoskeleton sensor data obtained during human movement tasks. The goal is to develop and validate predictive models that map ultrasound-derived muscle features to joint kinematics and neuromuscular function, enabling future real-time, closed-loop applications in rehabilitation robotics and gait training.

The student will develop and evaluate machine learning models that predict lower-limb kinematics from ultrasound data, using motion capture and EMG as ground truth references. These algorithms will ultimately be extended for real-time implementation, supporting closed-loop control strategies in which ultrasound-based muscle state estimation informs robotic or assistive device behavior.

With mentorship throughout the project, the student will:

1. Work with existing datasets including ultrasound imaging, motion capture, EMG, and exoskeleton sensor data collected during human movement tasks.
2. Preprocess and curate multimodal data streams, including processing of ultrasound images of muscles and their synchronization with biomechanical and electrophysiological signals.
3. Develop machine learning models (e.g., regression, deep learning, or time-series-based approaches) to predict joint kinematics and/or muscle function from ultrasound measurements.
4. Evaluate model performance using biomechanical outcome measures such as joint angles, movement timing, and muscle activation patterns.
5. Compare ultrasound-based predictions against motion capture and EMG data to assess model accuracy, robustness, and generalizability.
6. Contribute to the design of future real-time algorithms in which ultrasound-based predictions drive feedback or control of rehabilitation technologies.
7. Synthesize results through data visualization, technical reports, and potential contributions to manuscripts or conference presentations.

The student may also gain experience in real-time signal processing and control, including implementation of algorithms on microprocessor-based systems (e.g., Raspberry Pi or similar platforms) for future closed-loop applications.

Environment:

The student will be embedded in a highly interdisciplinary research environment that includes engineers, clinicians, physical and recreational therapists, and orthotists studying movement disorders and innovative rehabilitation technologies. Ongoing research spans rehabilitation robotics, wearable sensors, exoskeletons, musculoskeletal ultrasound, and mobile neuroimaging (EEG/fNIRS). The laboratory is housed in the NIH Clinical Center, the world’s largest hospital dedicated exclusively to clinical research, located on the main Bethesda campus. Students will have exposure to a wide range of career paths and training levels, including postbaccalaureate fellows, postdoctoral researchers, staff scientists, and clinician-investigators.