Microsoft, Providence & UW Release AI Tool to Speed Up Advanced Tumor Analysis

The tool, called GigaTIME, uses artificial intelligence to generate immune profiling data from standard pathology slides, insights that would otherwise require costly and time-intensive multiplex immunofluorescence (mIF) tests.
Pacific Northwest cancer researchers and technologists have released a new AI model that promises to make sophisticated tumor microenvironment analysis significantly faster and more affordable.
The tool, called GigaTIME, uses artificial intelligence to generate immune profiling data from standard pathology slides, insights that would otherwise require costly and time-intensive multiplex immunofluorescence (mIF) tests.
Developed by teams at Microsoft, Providence, and the University of Washington, the model is being publicly released along with a peer-reviewed study published in Cell. The tool is available on Hugging Face, GitHub, and Microsoft Foundry.
The model is designed to virtually recreate complex immune markers from hematoxylin and eosin (H&E) slides. These reconstructed features provide information about whether immune cells in the tumor microenvironment are actively responding to cancer—data that typically takes days of laboratory work and several thousand dollars per sample to obtain.
“Just for one sample, mIF analysis could easily take days and cost thousands of dollars,” said Hoifung Poon, general manager of Microsoft Research’s Real-World Evidence program. GigaTIME aims to bridge that gap by allowing clinicians and researchers to extract this information in a fraction of the time.
The initiative also builds on an earlier model, GigaPath, released last year for cancer diagnosis. Together, the efforts reflect the region’s increasing investment in applying AI to large-scale biomedical datasets.
The scale of GigaTIME’s development is among the largest reported for tumor microenvironment modeling. Researchers trained the tool on 40 million cells from Providence datasets, pairing pathology slides with mIF data covering 21 proteins. They then applied the model to samples from 14,256 cancer patients across 51 hospitals and over 1,000 clinics, generating a virtual library of nearly 300,000 mIF images spanning 24 cancer types and 306 subtypes.
“GigaTIME is about unlocking insights that were previously out of reach,” said Dr. Carlo Bifulco, chief medical officer of Providence Genomics.
Researchers say future work may integrate radiology, biopsy, and other clinical data to build more comprehensive models capable of predicting disease progression or treatment response. Such tools could eventually reduce the time and cost required to design clinical trials by enabling more informed patient selection and study planning.