STANFORD, Calif. – A new artificial intelligence model can predict a patient’s prognosis and response to immunotherapy for various cancers with remarkable accuracy, simply by analyzing standard hospital pathology slides. The breakthrough, detailed in a new study, promises to unlock a wealth of information hidden within routine tests, potentially revolutionizing cancer care.
AI Turns Routine Slides into Powerful Predictors
Researchers have developed a system called HEX that can virtually recreate detailed protein maps of tumors from commonly used hematoxylin and eosin (H&E) stained slides. This means doctors could soon gain deeper insights into a patient’s cancer without the need for costly and time-consuming specialized tests.
The ability to predict how a cancer will behave and respond to treatment is crucial for personalized medicine. Currently, this often relies on analyzing the expression of specific proteins within the tumor, a process that typically requires advanced techniques like immunohistochemistry (IHC). HEX offers a potentially faster, cheaper, and more accessible alternative.
“We’ve shown that a wealth of information about a tumor’s biology is already present in the H&E stain, which is routinely collected for all cancer diagnoses,” explained a researcher involved in the study. “Our model can unlock that information, providing clinicians with a more comprehensive picture of the cancer and helping them make more informed treatment decisions.”
How Does It Work?
The researchers trained HEX using data from over 2,300 patients with non-small cell lung cancer (NSCLC) and a broader pan-cancer dataset encompassing 34 different tissue types. The model learned to associate patterns in H&E images with the expression levels of 40 different proteins, as measured by a more complex technique called CODEX. HEX was then able to predict protein expression from H&E images alone.
To validate its accuracy, the team tested HEX on independent datasets, including those from the National Lung Screening Trial (NLST), The Cancer Genome Atlas (TCGA), and the PLCO Cancer Screening Trial. The results showed that HEX’s predictions closely matched the actual protein expression levels measured by CODEX and were able to accurately predict patient outcomes, including survival and response to immunotherapy.
The study involved analyzing H&E-stained slides from six cohorts – NLST, TCGA, PLCO, Stanford-TMA, TA-TMA, and Stanford immuno-oncology (Stanford-IO). Data on recurrence-free survival (RFS), disease-specific survival (DSS), overall survival, objective response, and progression-free survival (PFS) were analyzed for 2,150 patients, with an additional 148 patients assessed for immunotherapy response.
Further analysis extended HEX’s predictive capabilities to 12 additional cancer types within the TCGA database, involving over 5,000 patients. This retrospective study received approval from the Stanford University Institutional Review Board.
Beyond Lung Cancer: A Broadly Applicable Tool
The researchers also demonstrated that HEX could be used to predict prognosis across 12 additional cancer types, suggesting its potential for broad application in oncology. The model’s ability to generalize to different tissue types and staining protocols highlights its robustness and versatility.
The team also developed a framework called MICA to integrate the virtual protein maps generated by HEX with the original H&E images. This combined approach further improved the accuracy of predictions, particularly for immunotherapy response.
Looking Ahead
While HEX shows immense promise, the researchers emphasize that it is not intended to replace traditional diagnostic methods. Instead, it is envisioned as a complementary tool that can provide clinicians with additional information to guide treatment decisions. Further research is needed to validate HEX in larger clinical trials and to explore its potential for personalized cancer care.
The study’s findings represent a significant step forward in the field of computational pathology and could pave the way for a new era of AI-powered cancer diagnostics and treatment.
