The model has been trained on millions of clinical images and reports, annotated by medical specialists, ensuring it is the most capable specialized tool in radiology.

This launch comes at a crucial time, as the healthcare sector faces a shortage of radiologists, with only 11 per 100,000 people in the U.S., amidst rising imaging volumes.

Harrison.ai has unveiled Harrison.rad.1, a specialized vision language model designed to enhance radiology and improve healthcare access and patient outcomes.

Harrison.rad.1 can engage in open-ended discussions about X-ray images, detect and localize findings, and generate reports that consider clinical history.

In benchmark tests, the model scored 51.4 out of 60 on the Fellowship of the Royal College of Radiologists (FRCR) 2B exam, matching the performance of experienced radiologists.

Harrison.rad.1 significantly outperformed major large language models, achieving approximately double the score of competitors like OpenAI's GPT-4o and Google's Gemini-1.5 Pro.

The model also achieved 73% accuracy on RadBench, the highest among its peers, and 82% accuracy on closed questions in the VQA-Rad benchmark.

Prioritizing clinical safety and accuracy, Harrison.rad.1 builds on the responsible approach established by the existing solution, Annalise.ai, which is clinically cleared in over 40 countries.

Harrison.ai aims to tackle regulatory challenges in the U.S. that have created an innovation gap in AI for diagnostic medicine.

Harrison.ai has raised $141 million in funding and formed partnerships with major healthcare organizations, including Sonic Healthcare and I-MED Radiology Network.

The model is being made accessible to select industry partners, healthcare professionals, and regulators to encourage discussions on responsible AI use in healthcare.

Board Director Robyn Denholm described Harrison.rad.1 as transformative, marking a significant advancement for the company in addressing real-world healthcare challenges.