Microsoft has released several multimodal medical imaging foundation models as part of an expanded portfolio of healthcare AI offerings.
Now available in the Microsoft Azure AI model catalog, the models are designed for healthcare organizations to test, fine-tune, and build AI algorithms geared to their specific needs, according to the firm.
The models include the following:
- MedIimageInsight: An “embedding” model to support image analysis, including classifying and searching for similar medical images. Adapters can be utilized to streamline workflows in radiology and other image-generating specialties, enabling, for example, researchers to build models to automatically route imaging scans to specialists, or flag potential abnormalities for further review, Microsoft said.
- MedImageParse: This model provides image segmentation for imaging modalities such as x-rays, CT, MRI, and ultrasound.
- CXRReportGen: This multimodal AI mode incorporates current and prior images as well as key patient information to generate structured reports from chest x-rays. AI-generated findings are highlighted on the image, Microsoft said.
Microsoft said the models were developed in collaboration with Microsoft Research and strategic partners. The company noted that these foundation models could be used independently, connected to different modalities, or integrated with general reasoning models such as GPT-4 and Microsoft’s Phi family.
In other news, Microsoft has formally incorporated support for healthcare data – including medical images – in its Fabric analytics platform. The new capability enables organizations to integrate and utilize healthcare data for analytics, data science, and business intelligence needs, according to the vendor.
