Harrison.ai's Radiology Foundation Model Achieves Breakthrough Results in Rigorous Independent US Healthcare AI Challenge

Harrison.ai, a leading healthtech company, has announced that its radiology-specific foundation model, Harrison.rad.1, delivered outstanding results in an independent, large-scale evaluation conducted by Mass General Brigham and the American College of Radiology Data Science Institute. The findings highlight a major leap forward in AI-powered radiology reporting. Harrison.rad.1, released in 2023, outperformed foundation models from OpenAI, Anthropic, Google, and others on VQA-Rad, a widely recognized benchmark for evaluating multimodal AI models in medical contexts. The model achieved 82% accuracy and precision on closed questions related to plain radiographs, a performance now further validated by the recent Healthcare AI Challenge. The challenge tested the ability of AI-generated chest radiograph reports to mimic human radiologists in a Turing-style assessment. In blinded evaluations, 113 board-certified radiologists reviewed 2,840 reports across 117 cases during the 2025 ACR Annual Meeting. Harrison.rad.1’s AI-generated reports were deemed acceptable by 65.4% of radiologists—compared to 79.6% for human-written reports. While still below human performance, this result underscores rapid progress in AI draft report generation. Mass General Brigham noted that these findings reflect how fast AI solutions are advancing and their growing potential to enhance radiologist efficiency through automated draft reporting. The results come at a critical time for healthcare, as imaging volumes rise, radiologist shortages persist, and diagnostic backlogs grow. AI tools like Harrison.rad.1 could help address these challenges while maintaining high clinical standards. Dr. Aengus Tran, CEO and Co-Founder of Harrison.ai, expressed pride in the achievement. “This independent validation is exactly what’s needed to responsibly advance AI in radiology. While this is a significant milestone, our work continues. We’re committed to refining the model, expanding testing, and working with regulators to meet the clinical evidence required for real-world deployment.” The model’s success stems from its specialized training on millions of DICOM images and radiology reports across all X-ray modalities. Unlike general-purpose AI systems, Harrison.rad.1 is engineered for factual accuracy and clinical precision in radiology. Harrison.rad.1 is built on the same high-quality data and infrastructure that powers Harrison.ai’s clinical decision-support system, Annalise Enterprise CXR. This tool detects up to 124 chest X-ray findings and has been shown to improve diagnostic accuracy by 45% when used as an assistive tool for radiologists. Dr. Jarrel Seah, Chief AI Officer at Harrison.ai, emphasized the synergy between the foundation model and the company’s clinical tools. “Our AI algorithms are trained on some of the largest and most diverse hand-annotated radiology datasets in the world. This robust foundation enables accurate and reliable report generation.” Harrison.ai’s solutions have gained global regulatory recognition, with approvals in over 40 countries and deployments in 15. In Australia, its tools are accessible to 50% of radiologists. In England, they process more than 35% of chest X-rays. In the U.S., Annalise.ai has received FDA 510(k) clearance and Medicare NTAP status for multiple findings. Harrison.ai is a global healthtech company developing AI-powered diagnostic and workflow solutions. Its mission is to scale healthcare capacity, improve clinician support, and deliver better patient outcomes. The company’s platform powers multiple solutions, including Annalise.ai for radiology and Franklin.ai for pathology.