Voio, a frontier AI lab dedicated to healthcare, emerged from stealth with $8.6 million in seed funding from Laude Ventures and The House Fund. The company’s first focus is building a unified reading platform that helps radiologists across every scan and modality, spinning out from research labs at the University of California, Berkeley and University of California, San Francisco (UCSF). Voio’s founding team today released Pillar-0, an open-source AI model that interprets medical images directly to recognize hundreds of conditions from CT and MRI scans with unprecedented accuracy. The initial iteration of Pillar is the world’s most accurate AI model for medical imaging with a demonstrated 10% – 17% accuracy improvement over leading models from Google, Microsoft, and Alibaba.
Voio was founded by Adam Yala, Assistant Professor of Computational Precision Health at UC Berkeley and UCSF, Dr. Maggie Chung, Assistant Professor in Radiology and Biomedical Imaging at UCSF and a practicing radiologist, and Trevor Darrell, Professor of Computer Science at UC Berkeley and founder of Berkeley AI Research (BAIR). The team’s previous AI models have been validated in 92+ hospitals across 30 countries worldwide and their breast cancer tool has been used in over 2 million mammograms around the world.
Radiologists can drive better patient outcomes across the entire healthcare system: faster and more accurate diagnoses, proactive clinical management based on subtle imaging cues, and better patient access to the best of care. Voio aims to empower radiologists with state-of-the-art technology to push the clinical frontier, creating a higher standard of care and addressing the critical workforce shortage in radiology. With 375 million CT scans performed annually, this capacity gap is leading to longer turnaround times, increased burnout, and diagnostic delays that negatively impact patient care.
Currently, radiology reporting requires constant context-switching between the image viewer, reporting software, EHR, and AI tools. This fragmentation takes time away from image interpretation and fuels burnout. Voio restores balance with a unified reading environment powered by frontier vision-language models. Voio’s models interpret complete exams and draft high-quality reports so radiologists can review and finalize faster, without sacrificing accuracy.
“Radiologists shouldn’t have to choose between speed and quality,” said Dr. Maggie Chung, Co-Founder and Medical Lead of Voio. “Our goal is to make radiology reporting seamless by drafting full reports and connecting images, history, and prior exams into one intelligent platform that feels natural to use. By re-designing the reporting experience from the ground up, we can reduce unsatisfying grunt work and let radiologists focus on patient care.”
Pillar-0, the team’s open source model, achieved a .87 AUC across 350+ findings in chest CT, abdomen CT, brain CT, and breast MRI scans. This benchmark outperforms all publicly available AI models for radiology on the same test data, including Google’s MedGemma (.76 AUC), Microsoft’s MI2 (.75 AUC), and Alibaba’s Lingshu (.70 AUC). Moreover, P0 can easily be extended to tackle frontier clinical challenges; by fine tuning P0, the team improved over the state of the art in predicting future lung cancer, Sybil-1, by 7% in an external validation study at Massachusetts General Hospital. The team is now scaling the technology to support multi-modal agentic workflows in radiology and push preventative care.
Also Read: ARC, Mount Sinai Partner with NVIDIA to Decode Genome via AI
“The Voio team is showing that AI research can change healthcare outcomes. This work represents a fundamental shift in healthcare from reactive diagnosis to predictive medicine,” said Andy Konwinski, Co-Founder of Laude Ventures, Databricks, and Perplexity. “The ability to identify future health risks from current imaging before symptoms appear could transform how we approach preventive care, making radiology central to proactive health management rather than just documenting what’s already wrong.”
The technology builds on the teams’ proven track record of translating AI research into clinical practice. Dr. Yala previously created Mirai, a breast cancer risk prediction model validated across 2 million mammograms, and Sybil, which predicts lung cancer risk from screening scans. Dr. Maggie Chung has led a prospective study demonstrating how AI can reduce diagnostic workup time for high-risk breast cancer patients. Dr. Trevor Darrell founded Berkeley’s BAIR lab and led the team which built the Caffe deep learning framework which revolutionized image processing.
“Voio is the culmination of nearly a decade pushing the frontier of AI for health, building leading models for cancer imaging and validating them globally,” said Adam Yala, Co-Founder and CEO of Voio. “With Voio, we’re scaling our impact across radiology; we’re building a frontier AI lab and a unified reading platform that supports radiologists across every study and task. Over time, this foundation will enable richer AI systems that collaborate seamlessly across modalities and specialties.”
“Adam, Maggie, and Trevor represent exactly what we look for at The House Fund – domain expert Berkeley founders combining world-class AI research with proven clinical impact. Voio is building the infrastructure layer that will define the next generation of radiology, and we’re thrilled to support their mission,” said Jeremy Fiance, Founder of The House Fund in Berkeley.
Voio‘s open-source approach enables academic researchers to validate and build upon its models, creating transparency in a market where performance claims are often unverifiable. The company plans to support the development of independent benchmarks to establish evidence-based standards for radiology AI.
Source: Businesswire
