by
Gus Iversen, Editor in Chief | December 10, 2025
Researchers from UC Berkeley and UCSF have introduced Pillar-0, an open-source artificial intelligence model designed to analyze full 3D medical imaging volumes and detect hundreds of conditions across modalities including CT and MR.
The model is being positioned as a general-purpose foundation for medical imaging AI, with performance benchmarks that surpass current publicly available tools.
Unlike most existing AI systems that analyze 2D slices in isolation, Pillar-0 was built to interpret entire 3D imaging volumes directly. According to its developers, this enables the model to perform faster and more accurately while reducing the need for extensive retraining. The research team reported that Pillar-0 outperformed models from Google, Microsoft, and Alibaba by over 10% across 366 diagnostic tasks, achieving an area under the curve (AUC) of 0.87 on a UCSF data set.
Ad Statistics
Times Displayed: 357
Times Visited: 1 Stay up to date with the latest training to fix, troubleshoot, and maintain your critical care devices. GE HealthCare offers multiple training formats to empower teams and expand knowledge, saving you time and money.
“Pillar-0 outperforms leading models from Google, Microsoft and Alibaba by over 10% across 366 tasks and four diverse modalities; Pillar-0 also runs an order of magnitude faster, fine-tunes with minimal effort, and drives large downstream performance gains,” said Adam Yala, assistant professor at UC Berkeley and UCSF and senior author of the study.
The model’s architecture, known as Atlas, is designed to handle the computational complexity of 3D imaging. It processes abdomen CT scans over 150 times faster than traditional vision transformers, according to first author Kumar Krishna Agrawal, a Ph.D. student at UC Berkeley.
As part of the release, the team has also made available an evaluation framework called RaTE, which is intended to offer a clinically grounded benchmark for assessing AI performance on radiologist-relevant tasks. “We assembled a large collection of diagnostic questions and findings that radiologists routinely evaluate in clinical practice,” said Dr. Maggie Chung, assistant professor in radiology at UCSF and co-senior author.
The Pillar-0 codebase, pretrained models, and evaluation tools are available for public use, with ongoing plans to expand the model’s capabilities across additional modalities and use cases.
