Written by
Share article:
Harrison.ai has unveiled a multimodal LLM, which has been designed to accomplish radiology tasks.
The model, called Harrison.rad.1, is dialogue-based and accepts interleaved text and visual inputs, and can generate both structured and unstructured text outputs. The model has been trained on real-world data, and, according to the vendor, has been designed with a focus on factual correctness and clinical accuracy.
Why we are covering this topic
– Harrison.AI is one of the leading AI vendors, and its move into generative AI marks a significant step.
– There are many questions about the optimum usage of generative AI in radiology.
Why it matters
– Harrison, through its radiology venture Annalise, is gaining considerable traction, recently being made available to UK NHS hospitals, and being granted Medicare New Technology Add-on Payments in the US. This traction could make it a potent force in commercialising generative AI in radiology.
– Amidst a shortage of radiologists, tools which are able help clinicians could be valuable, but only if they can be trusted.
– Given the vendor compares the scores of its model to the scores of radiologists undertaking a radiology certification examination, it raises questions about the role of gen AI models alongside doctors.
