Clinical Translation of AI in Radiology

Many powerful radiology AI tools have been developed and FDA-approved, yet relatively few live up to the performance and promise demonstrated on paper. The central question for our lab is what happens when AI reaches the clinic. We study how to implement AI effectively and safely in clinical practice and how to continuously monitor deployed AI systems so that their performance lives up to their promise and their safety holds up over time as models, data, and workflows change.

We are a deeply collaborative team. Together with computer scientists, cognitive scientists, and clinicians across both academic and industry, we build AI tools and infrastructure surrounding them so that they understand clinical context, work across the full radiology workflow, and remain safe and reliable after deployment. Current tools we work with include clinical history summarization, automated imaging protocol selection, intelligent radiology report revision agents, automated resident quality assurance, automated rad-path-clinical discrepancy detection, and various computer vision models for cardiothoracic imaging. We evaluate each by its real-world effects on clinical efficiency, accuracy, and patient safety.

Our research spans both the technical foundations of model reliability and uncertainty and strategies for effective human-AI collaboration. This includes simulation environments for realistic training and evaluation, analysis of human-AI collaborative behavior, and learning methods for multimodal reasoning across complex clinical workflows. Together, our goal is to accelerate the clinical translation of AI into radiology while ensuring its safety and effectiveness. (last updated: October 2026)

AI for Pulmonary Imaging

Our lab combines clinical expertise in lung cancer and interstitial lung disease (ILD) with modern AI to improve how pulmonary disease is detected, characterized, and managed. Our work grew out of quantitative imaging, and today we focus on AI systems that interpret lung imaging in light of each patient’s clinical context, supporting diagnosis, risk stratification, and management decisions where they matter most. We develop and validate in-house algorithms, partner with industry to translate advanced imaging and machine learning tools into practice, and evaluate how these tools perform alongside radiologists and referring clinicians. 

We collaborate closely with the lung cancer screening programs and tumor boards at UCSF, ZSFG, and SFVA. Current work includes AI for pulmonary subsolid nodule risk stratification that integrates imaging with patient history, decision support for nodule management, and patient-centered communication of screening results. We also closely partner with the multidisciplinary UCSF Interstitial Lung Disease Program to develop imaging biomarkers and AI tools for early detection and prognostication of ILD (e.g. interstitial lung abnormality or mild fibrotic changes seen in CT). (last updated: October 2026)

Clinical Translation of Mid-Field (0.55T) Lung MRI

We closely collaborate with UCSF advanced imaging science teams to support the clinical translation of cutting-edge hardware and software innovations for cardiothoracic imaging. We have helped to establish UCSF as one of the most active centers in the nation for the pulmonary application of 0.55T Lung MRI. 

News coverage here. (last updated: October 2026)