The Anatomical Foundation Model (AFM) is learning to parse and identify all structures in a whole body scan based on their shapes and locations in the body – mirroring the foundations of a radiologists training. This approach means the AFM can be applied to other imaging modalities and can derive measurements that are less impacted by the subject breathing or moving, making them useful for longitudinal analyses unlike vision based approaches.

Once this foundation is built, the AFM will learn how to identify low risk scans without abnormalities. It will do this by detecting regions of interest, analyzing tissue properties and considering the broader Gemini Digital Twin context including prior scans, lab results, genetics, family and medical history. This will enable the intelligent routing of scans that need further review to the appropriate radiologist, making optimal use of their time and equipping them with quantitative data and insights to aid their interpretation. The AFM will learn from those interpretations to improve its routing and tune its analyses and outputs for specific diseases.

The AFM combines the pattern recognition that humans apply to interpretation with an ability to reason over more quantitative data than any single person could process at once. When applied to the multi-parametric quantitative data provided by Mark I’s TFM technology, the AFM’s analyses will be independent of image contrasts, features and textures, dramatically increasing their accuracy and unlocking truly consistent longitudinal tracking. In the context of the highly accessible recurring exams we envision, these quantitative and reproducible outputs that the AFM + Mark I provide will address the false positive concerns that exist with one off whole body MRIs today.