Microsoft just released two distilled pathology models that cut compute cost by 50x. The models are open source, they fit on a single GPU, and they make population-scale cancer research possible for teams that could never afford it before.
Press enter or click to view image in full size
Use this link to access this for free :https://inamdaraditya.medium.com/microsoft-just-made-pathology-ai-50x-cheaper-here-is-why-that-changes-cancer-research-712d9b66c894?sk=d884159c88ac6a060ea5c8b7561db67b
The problem with computational pathology was never whether the models could find the biology. They could. The problem was whether anyone could afford to run them on enough patients for the findings to mean anything.
A single cancer cohort study can require millions of tile inferences across tens of thousands of slides. At the cost of the original GigaPath model, that study was out of reach for everyone except the largest research labs. Last month, Microsoft released two models that cut the compute cost by roughly 50x while keeping 97% of the performance.
They are called GigaPath-Flash and GigaTIME-Flash. They are open source. And they change what kind of…