Stanford HAI Grants Fund Innovative POCUS Training with AI-Enhanced Datasets

The field of medical education is about to get a significant boost thanks to a seed research grant from the Stanford Institute for Human-Centered Artificial Intelligence (Stanford HAI). The project, co-funded with the Stanford Center for Digital Health, focuses on improving training for point-of-care ultrasound (POCUS) using AI-enhanced datasets. POCUS allows clinicians to perform real-time assessments at the patient's bedside, but current training methods are often inconsistent and resource-intensive.

The primary investigator, Dr. Andre Kumar from the School of Medicine’s Department of Hospital Medicine, aims to create a first-of-its-kind database that will address these challenges. The project will collect and annotate 7,500 ultrasound clips covering heart, lung, and abdominal imaging-three of the most common clinical applications of POCUS. Unlike existing datasets that focus primarily on disease detection, this new database will include dual annotations: one for measuring the quality of image acquisition and another for identifying potential pathology.

Enhancing POCUS Training with Dual Annotations

The unique aspect of this project lies in its dual annotation approach. Each ultrasound clip will be annotated not only to identify potential pathologies but also to assess the quality of the acquisition. This means that common errors in technique, such as improper probe placement or insufficient gel application, will be flagged. By doing so, the dataset can serve as a training tool that simulates expert faculty guidance, providing real-time, personalized feedback to learners.

• Data Collection: The team will gather 7,500 ultrasound clips from Stanford Medical Center.
• Annotation Process: Each clip will be annotated for both image quality and potential pathology.
• AI System Development: The dataset will be used to develop AI systems that can provide real-time feedback to trainees.

This dual annotation approach is crucial because it addresses a significant gap in current POCUS training. Many medical students and residents learn with minimal supervision, which can compromise patient care. By providing immediate feedback on both the quality of the image and any potential issues, the AI system will help trainees improve their skills more effectively.

What to Watch

The implications of this project extend beyond just improving POCUS training. The publicly released database has the potential to accelerate medical education research globally. Future applications could include educational dashboards that track trainee progress and provide personalized learning paths. These tools could transform how ultrasound is taught, ensuring that AI augments rather than replaces human clinical judgment.

Dr. Kumar’s project also addresses broader issues in the field of learning sciences. Historically, there has been a significant gap between research and practice, making it challenging for practitioners to obtain actionable evidence for instructional methods that apply to their unique contexts. By creating a robust, annotated dataset, this project could help bridge that gap, providing a foundation for more rigorous testing of instructional methods.

The Fall 2026 cohort for Stanford HAI and the Stanford d.school’s Human-Centered AI for Social Impact program will further contribute to these efforts by fostering interdisciplinary collaboration and innovation in AI applications. With initiatives like Dr. Kumar's POCUS project, the future of medical education looks promising, as AI continues to play a crucial role in enhancing both teaching and patient care.