AI startup Hoppr teamed up with AWS to launch a new foundation model to help bring more generative AI solutions into medical imaging, the companies announced on Sunday at RSNA 2023, the annual radiology and medical imaging conference in Chicago.
The new product, named Grace, is a B2B model designed to help application developers build better AI solutions for medical images — and to build them more quickly. Along with the launch of Grace, Hoppr also announced that it received “a multi-million dollar investment” from Health2047, the American Medical Association’s venture studio.
Unlocking Transparency in PBM Pricing
The TSX Venture Exchange has a strong history of helping early-stage health and life sciences companies raise patient capital for research and development.
Chicago-based Hoppr, which was founded in 2019, has raised $4.1 million to date, said CEO Khan Siddiqui. Pioneering computer scientist Grace Hopper is the namesake for both the company and its new product, he noted.
The foundation model is meant to enable “image-to-image and text-to-image learning” across all medical imaging modalities, Hoppr said in its press release.
“Grace is trained on a very large data set that enables the foundation model to learn across various imaging modalities and radiology reports. We use embedding and vectors from the training on the images to map and correlate the same findings in different modalities. This allows Grace to show how a nodule in an x-ray would be represented in a CT, for example,” Siddiqui explained.
The model can provide diagnostic, clinical and operational insights from medical imaging data, he added. For example, Grace can let users know when additional diagnostic imaging is needed at the point of care, and the model can provide findings from an image to prepopulate a preliminary report that the clinician can finalize after review.
When Investment Rhymes with Canada
Canada has a proud history of achievement in the areas of science and technology, and the field of biomanufacturing and life sciences is no exception.
The model can also help an interventional neurologist better plan a procedure by allowing them to converse with medical imaging studies, Siddiqui said. They can ask questions about findings, alternative imaging views, treatment protocols and recommended surgical interventions.
Grace is currently available via an API service for use by application developers. Hoppr’s vision is that these developers will use the model to create products that allow radiologists, technicians and other staff members to engage with medical images more meaningfully.
The model could help developers create AI solutions more quickly because the current approach to application development — supervised learning — requires them to see what is being annotated in each image and takes a year or more, Siddiqui pointed out. With Grace, developers can fine-tune their model and wait “just weeks” to get to market, he said.
Hoppr built Grace exclusively on AWS using Amazon SageMaker, the tech giant’s machine learning platform designed to facilitate the development and deployment of cloud-based AI models. The quality of AWS’ technology and data storage, as well as its ability to scale, made the cloud service “a natural home” for Hoppr, Siddiqui explained.
“The company’s natural and deep understanding of medical imaging is unparalleled,” he declared.
Photo: metamorworks, Getty Images