How a New Pathology Foundation Model Reads Tumour Blueprints

Imagine you are looking at a satellite photograph of a sprawling city. You can easily spot tall buildings, winding roads, and wide green parks. However, you do not know the rules that built them. You cannot see the architectural blueprints. You cannot read the local zoning laws. If a road suddenly curves, you might guess it is to avoid a hill, but you cannot be sure without a map. Standard computer systems that look at medical images do exactly this. They look at the physical shapes of cells, but they miss the genetic instructions that created those shapes.

Now, imagine a highly advanced software system. This system has studied millions of satellite photos side-by-side with the actual city blueprints. If it spots a specific cluster of tall buildings, it knows exactly which zoning law caused it. If it sees a curved road, it knows the exact geological fault line underneath. This is the exact logic behind a new artificial intelligence system called Fuji.

Building a Smarter Pathology foundation model

Scientists wanted to connect the physical appearance of tissue directly to its genetic code. To achieve this, they built Fuji, a highly advanced pathology foundation model. The researchers fed this system over 60,000 highly detailed images of tissue samples. But they did not stop at pictures. They paired these images with tens of thousands of DNA and RNA profiles from major biological libraries. By learning from both at once, the model learned to match the visual patterns on a standard medical slide with the hidden genetic mutations that caused them.

Let us break down how this works step-by-step. If a cell's DNA becomes unstable, it changes how the cell grows and divides. Then, those physical changes alter the overall architecture of the tissue. Finally, these changes appear on a standard glass slide under a microscope. Fuji works backwards through this chain of events. It looks at the shapes and colours on the slide and predicts the genetic errors. It acts as a direct translator between what we can see and what is hidden deep inside the genome.

Spotting Hidden Signs of Disease

The research team measured how well Fuji could spot genetic issues compared to older models. The results were highly accurate. Fuji successfully identified major genetic events, such as missing DNA repair mechanisms and whole-genome duplication. It even spotted the physical footprint of extrachromosomal DNA. These are rogue pieces of genetic material that sit outside the normal chromosomes and often help a tumour resist treatment. The model could also map out where damage from tobacco exposure occurred within the tissue architecture.

What does this mean for the future of medicine? Currently, finding these genetic details requires expensive and slow DNA sequencing. Because Fuji can read these signs from a standard, cheap tissue slide, it suggests that advanced tumour profiling could become much faster and more widely available. It may help doctors quickly identify which patients will respond to specific treatments, like immunotherapy. By linking physical tissue shapes to genetic blueprints, this tool offers a clear new method to understand and treat disease.