New York, NY [July 9, 2025]—A new study by researchers at the Icahn School of Medicine at Mount Sinai, Memorial Sloan Kettering Cancer Center, and collaborators, suggests that artificial intelligence (AI) could significantly improve how doctors determine the best treatment for cancer patients—by enhancing how tumor samples are analyzed in the lab.
The findings, published in the July 9 online edition of Nature Medicine https://www.nature.com/articles/s41591-025-03780-x , showed that AI can accurately predict genetic mutations from routine pathology slides—potentially reducing the need for rapid genetic testing in certain cases.
"Our findings show that AI can extract critical genetic insights directly from routine pathology slides," says study lead author Gabriele Campanella, PhD, Assistant Professor of the Windreich Department of Artificial Intelligence and Human Health at the Icahn School of Medicine at Mount Sinai. "This could streamline clinical decision-making, conserve valuable resources, and accelerate patients' access to targeted therapies by reducing reliance on certain rapid genetic tests."
Using the largest dataset of lung adenocarcinoma pathology slides matched with next-generation sequencing results from multiple institutions across the United States and Europe, the investigators set out to test whether AI could help streamline cancer care.
For patients with lung adenocarcinoma—the most common type of lung cancer—genetic testing known as somatic sequencing is a critical step. It detects mutations in the tumor's DNA that aren't inherited but instead develop over a person's lifetime. These acquired mutations guide doctors in selecting personalized treatments. But the tests can be expensive, time-consuming, and aren't always available, even at leading hospitals.
To explore a faster, more accessible option, the researchers trained their AI on H&E-stained pathology slides—the standard pink-and-purple tissue images pathologists use to diagnose cancer under the microscope. These slides are prepared from tumor samples collected during standard diagnostic biopsy and are a routine part of nearly every patient's diagnostic workup.
"We asked: could we train AI to predict genetic mutations using standard pathology slides, which are already part of every patient's workup?" Dr. Campanella says. "This could support faster treatment decisions—without compromising quality of care."
The team developed a novel AI model that fine-tunes large "foundation" models for a specific task—in this case, predicting EGFR (epidermal growth factor receptor) mutations from these slides. EGFR is a protein on cell surfaces that helps them grow and divide.
Mutations in the EGFR gene can drive cancer growth, especially in patients with lung adenocarcinoma. Identifying these mutations is critical because they make tumors highly responsive to targeted therapies—but only if detected. While confirmation still requires advanced genetic testing, researchers are exploring how AI could help flag likely cases earlier and more efficiently, making better use of limited tumor samples and accelerating the path to treatment.
In a real-time, behind-the-scenes "silent trial"—the first of its kind in pathology—the AI analyzed live patient samples at Memorial Sloan Kettering Cancer Center. The AI's predictions weren't visible to clinicians but showed that it could reliably detect EGFR mutations and potentially reduce the need for rapid genetic tests by more than 40 percent, the researchers say. To prove generalizability, data from hospitals in the United States and Europe was analyzed retrospectively.
"This study, which involved known biomarkers, shows how AI can be thoughtfully integrated into cancer diagnostics to support faster, smarter, and more personalized care," says Alexander Charney, MD, PhD , Vice Chair, Windreich Department of Artificial Intelligence and Human Health , and Associate Professor of Artificial Intelligence and Human Health, Psychiatry, Genetics and Genomic Sciences, and Neuroscience at the Icahn School of Medicine. "By flagging key mutations earlier, it helps oncologists act quickly—while also reducing the burden on sequencing labs in high-resource settings that run the rapid tests. The real promise lies not only in efficiency, but in the future potential to uncover new biomarkers from routine pathology slides. Rigorous, real-time trials like this one are exactly what we need to safely and responsibly bring AI into hospitals."
The team is continuing data collection through the silent trial and planning to expand it to additional sites, laying the groundwork for the regulatory approval process. Longer term, the research team aims to broaden the system's capabilities to detect additional cancer biomarkers and to evaluate its impact in lower-resource settings, where access to genetic testing is more limited. Together, these efforts can lead to broader clinical adoption of AI and improved patient outcomes in both low and high resource settings.
The paper is titled "Enhancing Clinical Genomics in Lung Adenocarcinoma with Real-World Deployment of a Fine-Tuned Computational Pathology Foundation Model."
The study's authors, as listed in the journal, are Gabriele Campanella, Neeraj Kumar, Swaraj Nanda, Siddharth Singi, Eugene Fluder, Ricky Kwan, Silke Muehlstedt, Nicole Pfarr, Peter J. Schüffler, Ida Häggström, Noora Neittaanmäki, Levent M. Akyürek, Alina Basnet, Tamara Jamaspishvili, Michel R. Nasr, Matthew M. Croken, Fred R. Hirsch, Arielle Elkrief, Helena Yu, Orly Ardon, Gregory M. Goldgof, Meera Hameed, Jane Houldsworth, Maria Arcila, Thomas J. Fuchs, and Chad Vanderbilt.
This work was supported in part by the AI-Ready Mount Sinai (AIR.MS) platform and the expertise of the Hasso Plattner Institute for Digital Health at Mount Sinai (HPI.MS). Computational resources and expertise were also utilized from Scientific Computing and Data at the Icahn School of Medicine at Mount Sinai, supported by the Clinical and Translational Science Awards (CTSA) grant UL1TR004419.
In addition, research funding was provided by a Cancer Center Support Grant from the NIH/NCI (P30CA008748) and the Warren Alpert Foundation through the Warren Alpert Center for Digital and Computational Pathology at Memorial Sloan Kettering Cancer Center.
