AI Tool Predicts Cancer Gene Signatures from Biopsy Images with High Accuracy

An artificial intelligence (AI) tool developed by Stanford Medicine researchers can accurately predict cancer (search)Search disease gene signatures from standard biopsy images. The AI, named SEQUOIA (slide-based expression quantification using linearized attention), analyzes hematoxylin and eosin (H&E)-stained biopsy images to predict the expression patterns of over 15,000 genes. This innovation could potentially reduce the need for expensive gene expression tests and provide a new source of data for cancer research and clinical decision-making.

The researchers trained SEQUOIA using 7,584 cancer (search)Search disease biopsies from 16 different cancer types, each prepared with standard H&E staining. Transcriptomic data, indicating which genes the cells were actively using, was also integrated. By combining these datasets with images and transcriptomic data from thousands of healthy cells, the AI program learned to predict gene expression patterns from the stained images.

High Correlation with Real Gene Activity

SEQUOIA demonstrated a high degree of accuracy in predicting gene activity. For some cancer (search)Search disease types, the AI-predicted gene activity had a correlation of over 80% with the actual gene activity data. According to Olivier Gevaert, PhD, professor of biomedical data science and biomedical informatics, the model's performance improved with the inclusion of more samples for a given cancer type. "It took a number of iterations of the model for it to get to the point where we were happy with the performance," Gevaert said. "But ultimately for some tumor types, it got to a level that it can be useful in the clinic."

Predicting Genomic Programs and Patient Outcomes

SEQUOIA is particularly adept at predicting the activation of large genomic programs, such as those related to inflammation or cell growth. The AI displays genetic findings as a visual map of the tumor biopsy, allowing scientists and clinicians to observe genetic variations in different tumor areas. To assess SEQUOIA's clinical utility, the researchers focused on breast cancer (search)Search disease genes already used in commercial genomic tests, such as the MammaPrint (search)Search drug test, which analyzes 70 breast-cancer (search)Search disease-related genes to assess recurrence risk.

"Breast cancer (search)Search disease has a number of very well-studied gene signatures that have been shown over the past decade to be highly correlated with treatment responses and patient outcomes," Gevaert explained. The team demonstrated that SEQUOIA could provide similar genomic risk scores to MammaPrint (search)Search drug using only stained images of tumor biopsies. These results were consistent across multiple groups of breast cancer patients, with patients identified as high risk by SEQUOIA experiencing worse outcomes, including higher recurrence rates and shorter time to recurrence.

Future Applications and Clinical Potential

While SEQUOIA is not yet ready for clinical use and requires further testing in clinical trials and FDA approval, Gevaert's team is actively improving the algorithm and exploring its potential applications. The researchers believe that SEQUOIA could significantly reduce the need for expensive gene expression tests in the future. "We’ve shown how useful this could be for breast cancer (search)Search disease, and we can now use it for all cancers and look at any gene signature that is out there," Gevaert said. "It’s a whole new source of data that we didn’t have before."

Scientists from Roche Diagnostics (search)Search company also contributed to the research.