Harrison.ai, a leader in healthcare AI technology, has introduced Harrison.rad.1, an advanced radiology-specific vision language model. This breakthrough aims to address global healthcare challenges by offering enhanced AI capabilities for radiology, and is being made available to selected partners and professionals for evaluation.
Key Features of Harrison.rad.1:
- Dialogue-Based Model: Capable of open-ended chat regarding X-ray images, detecting and localizing radiological findings, and generating detailed reports.
- Clinical Safety and Accuracy: Prioritizes precision and safety in its applications, leveraging extensive clinical data for training.
- Unique Training Dataset: Trained on millions of annotated images and reports, Harrison.rad.1 uses proprietary clinical data to provide clinically accurate training signals.
Performance and Impact:
- Outperforms Major LLMs: Achieved approximately 2x better performance than major language models in the FRCR 2B exam.
- High Accuracy: Scored 51.4 out of 60 in the FRCR 2B exam, surpassing other foundational models that scored below 30.
- Benchmark Achievements: Excelled with 82% accuracy on closed questions in the VQA-Rad benchmark and 73% in RadBench, outperforming other models.
Leadership and Vision:
- CEO Quote: Dr. Aengus Tran emphasized the model’s groundbreaking nature and its commitment to high-quality, ethical AI development.
- Board Director Quote: Robyn Denholm highlighted the transformative potential of Harrison.rad.1 in solving real-world problems and saving lives.
Ethical and Collaborative Approach:
- Responsible AI Development: Building on the success of Annalise.ai, Harrison.rad.1 will undergo open evaluations by leading professionals.
- Accessibility: The model is accessible to researchers, industry partners, and regulators to foster conversations on the safe and responsible use of AI in healthcare.
Harrison.rad.1 represents a significant advancement in radiology AI, offering unprecedented accuracy and flexibility. The model’s performance in radiology examinations and benchmarks underscores its potential to revolutionize healthcare and improve patient outcomes.
