Work

A selection of my research and key projects, featuring the work I’m most excited about.

Publications: 11
Citations: 497
h-index: 7
i10-index: 5

Featured Work

arXiv preprint · 2026

Match4Annotate

Transferring point and mask annotations across independently acquired ultrasound videos without target-side labels.

Match4Annotate transfers user-defined point and mask annotations from a labeled ultrasound video to independently acquired target videos without target-side labels or manual initialization. The method combines continuous spatiotemporal DINOv3 feature fields with implicit feature flow, aligning anatomy in feature space rather than relying on ultrasound intensity. Across four cardiac and musculoskeletal datasets, it achieves state-of-the-art cross-video annotation transfer and can reuse annotations across datasets with different labeling protocols, without task-specific training.

Paper Code · coming soon

Experimental setup synchronizing reaching motion, ultrasound, motion capture, electromyography, and accelerometry

Scientific Data · 2026

Multimodal Biomechanics Dataset

Connecting internal tissue motion with synchronized kinematics, EMG, and accelerometry.

This open dataset synchronizes transverse upper-arm ultrasound with 3D motion capture, EMG, and accelerometry from 36 participants performing reaching movements. I annotated and tracked 11 anatomical points across nearly 300,000 frames, creating the first dataset to provide continuous frame-level point tracking in ultrasound videos. The release supports supervised training and benchmarking of ultrasound point-tracking models and includes code and tutorials for movement analysis.

Paper Data Code

Scientific Reports · 2026

DUSTrack

Accurate, semi-automated point tracking across diverse ultrasound applications.

DUSTrack is a semi-automated workflow that combines a fine-tuned ResNet-50 with Lucas-Kanade optical flow to accurately track arbitrary points in B-mode ultrasound videos. Its modular design allows the annotation, model, and postprocessing components to be adapted or improved independently, while the current implementation provides an accurate, general-purpose tracking system evaluated on cardiac wall motion, muscle deformation, and fascicle dynamics.

Paper Code

MIT 6.7920 · 2025

RL4UltraBot

Closing the loop between ultrasound perception and robotic control through reinforcement learning.

I trained a PPO policy to control a simulated three joint robotic arm and keep a moving bone in view in B-mode ultrasound. A perception model combining frozen DINO features with a trained detection head localizes the bone and provides visual feedback to the controller.

Report Code

Nature Medicine · 2023

Tissue Perfusion Pressure

A new hemodynamic measure for continuous monitoring and risk prediction in critical care.

We developed tissue perfusion pressure, a hemodynamic measure derived from routinely collected arterial blood pressure signals. I contributed to the data analysis and interpretation across 5,988 cardiac ICU admissions and an external cohort of 864, where the measure predicted mortality risk, hospital stay, and peak lactate, outperforming mean arterial pressure, the current clinical standard.

Paper

More Research & Publications

Peer-reviewed publications, preprints, conference contributions, projects, and academic theses.

* Equal contribution.

2026

A multimodal biomechanics dataset with synchronized kinematics and internal tissue motions during reaching

Roger Pallarès-López, Duarte Folgado, Uriel Magana-Salgado, Jessica Rosendorf, Enya Ryu, et al.

Scientific Data 13, 709Published

View Code

2026

DUSTrack: Semi-automated point tracking in ultrasound videos

Praneeth Namburi*, Roger Pallarès-López*, Jessica Rosendorf, Duarte Folgado, Brian W. Anthony

Scientific Reports 16, 13340Published

View Code

2026

Cohort-Scale Neural Atlases of Ultrasound Video

Zhuorui Zhang, Roger Pallarès-López, Xuan Wu, Praneeth Namburi, Brian W. Anthony

arXiv:2606.00890Preprint

View

2026

Match4Annotate: Propagating Sparse Video Annotations via Implicit Neural Feature Matching

Zhuorui Zhang*, Roger Pallarès-López*, Praneeth Namburi, Brian W. Anthony

arXiv:2603.06471Preprint

View