For many patients, the scariest part of undergoing treatment may be the feeling of claustrophobia while lying in a cramped, noisy tube during an MRI scan or waiting for CT scan results to reveal the progression of heart disease. Beyond the walls of the exam room, radiologists face another source of anxiety: an overwhelming and increasing workload.
Thanks in part to advances in medical imaging, healthcare providers are ordering more and more diagnostic images to better understand their patients’ health and avoid invasive procedures such as biopsies. At the same time, the world’s population is aging and more patients are presenting with conditions that require imaging tests. However, despite the rapidly increasing need, the number of radiologists has not kept up. This trend has accelerated during the coronavirus pandemic, with more radiologists leaving than usual. These trends are leading to physician burnout and delayed patient outcomes. According to the American College of Radiology, 2025 will be the third year in a row that labor shortages will be the biggest threat to the radiology field.
Researchers at the University of California, Berkeley and the University of California, San Francisco are using artificial intelligence to address this need. This is part of a growing trend in medicine to use AI to enhance the work of healthcare professionals while addressing challenges such as rising healthcare costs and disparities in access.
In 2025, researchers from Berkeley and UCSF launched Voio, a startup aimed at building AI models that help radiologists interpret images faster and more accurately. Voio’s tools are designed to help draft reports, freeing up radiologists to focus on the patient, predicting a patient’s risk for serious diseases like cancer, osteoporosis, and heart failure years in advance, and even predicting how an individual will respond to different treatment plans.
“We are empowering individual radiologists to have more impact, even with their heavy workloads, and ultimately save more patient lives,” said Voio CEO Adam Yara, assistant professor of computational precision health, statistics, and computer science at the University of California, Berkeley and the University of California, San Francisco. Voio plans to make similar advances through AI in other medical fields.
Yala launched Voio with co-founders Dr. Maggie Chung, a resident assistant professor in the UCSF Department of Radiology and Biomedical Imaging, and Trevor Darrell, a resident professor in the Department of Electrical Engineering and Computer Science at the University of California, Berkeley.
Before launching Voio, researchers developed Pillar-0, an open-source AI model trained on UCSF’s treasure trove of medical images to detect current conditions such as brain hemorrhages and immediate concerns that radiologists cannot detect, such as long-term lung cancer risk. According to Yala, Pillar-0 is the world’s best foundational AI model in radiology today. Teams of researchers, engineers, and doctors around the world are leveraging this to create better cancer prediction models and diagnostic tools.
Voio is currently developing Pillar-1. This is a new AI model that can detect patient risk associated with a variety of medical threats from a wider range of images and integrates the results into a draft report for radiologists. Yara says this will aid in the interpretation of the most complex cases and provide insight into disease progression that radiologists currently cannot detect. Pillar-1 is part of a system Voio is developing that can also perform tasks that don’t require specialized medical training in radiology, such as transcribing doctors’ voice notes and collating patient data.
Making radiology more efficient and accurate
Chung is excited to help her fellow radiologists spend more time caring for their patients. “By reducing non-technical manual tasks, we can bring back the joy of work to radiologists,” she said. “It brings us back to why we became radiologists. We are the detectives behind the scenes in the hospital. Through image processing, we derive important findings that have a big clinical impact on patients.”
Yala’s expectations for Voio go far beyond being a virtual work assistant. We hope that AI will revolutionize clinical guidelines for radiologists. “The way we think about public health has to change,” he says. “The way we deliver digital advertising cannot be more sophisticated and personalized than the way we deliver cancer screening.”
It’s an ambitious goal that Yara first conceived during his doctoral research at MIT, where he developed Mirai. Mirai is an open-source AI model that can identify people at high risk for breast cancer years before radiologists can. He later designed Sybil, an open source model that does something similar for lung cancer risk. Together, more than 90 hospitals in 30 countries are conducting research and testing using Mirai and Sybil, and in some cases are developing their own medical AI models based on them, Yara said. A recent Mirai prospective study led by Chung found that the use of AI may allow women at high risk for breast cancer to be evaluated more quickly. Several hospitals across the United States are recruiting patients for a new clinical trial to further study Mirai’s breast cancer detection rates.
“These tools are advancing the state of the art in oncology,” Yara said. “Being able to see into the future enables new types of clinical care. You can be proactive.”
Darrell said AI tools already exist to automate some of radiologists’ jobs, but they haven’t been proven to improve overall radiologist productivity. “That’s what really makes a difference,” Darrell said. “We no longer need automation to add bells and whistles in front of radiologists. We need AI to make radiologists more effective, accurate, and productive. That’s what we’re building.”
Yala, Chung, and Darrell’s collaboration grew out of the UCSF/UC Berkeley Joint Program in Computational Precision Health (CPH). CPH was founded in 2021 to apply a combination of technological approaches, including bioinformatics, genomics, machine learning, and simulation, to improve clinical care. Yara was one of CPH’s first faculty members. He still drives between campuses to teach. From the beginning, Yala embraced CPH’s goal of developing AI models that have a real impact on people’s lives.
“I think CPH has the best ecosystem in the world to enable this kind of innovation,” Yara said. “I’m so grateful to be here.”
Yara said that as a private startup, Voio had access to orders of magnitude more data than university teams to develop Pillar 1 and other models. “Voio is rapidly making huge leaps forward in what AI models can do,” he added.
Most of all, Yala said he’s excited to bring Voio’s AI tools into the hands of doctors, making a real difference for overwhelmed radiologists. “Our model and products will make it really more fun and empowering to be a radiologist next year than it was last year,” he said. In the future, it is hoped that radiologists will no longer have to worry about heavy workloads and will be able to focus more on patient care.
