New York, New York [July 9, 2025]- New research by ICAHN School of Medicine, Memorial Sloan Kettering Cancer Center in Mount Sinai, and collaborator researchers suggest that artificial intelligence (AI) can significantly improve the way physicians determine the best treatment for cancer patients by increasing the way tumor samples are analyzed in the lab.
Findings published in the July 9th online edition of Nature Medicine https://www.nature.com/articles/S41591-025-03780-x showed that AI can accurately predict genetic variation from daily pathology slides.
“Our findings show that AI can directly extract important genetic insights from everyday pathology slides,” said author Gabriele Campanella, PhD, assistant professor in the Faculty of Artificial Intelligence and Human Health at Icahn School of Medicine in Mount Sinai. “This allows for increased access to targeted therapies for patients by streamlining clinical decision-making, saving valuable resources, and reducing reliance on specific, rapid genetic testing.”
Using the largest dataset of lung adenocarcinoma pathology, using slides consistent with next-generation sequencing results from multiple institutions in the US and Europe, investigators set out to test whether AI can help streamline cancer care.
For patients with lung adenocarcinoma, the most common type of lung cancer, genetic testing, known as somatic cell sequencing, is an important step. Detects tumor DNA mutations. This is not inherited, but instead occurs throughout a person's life. These acquired mutations guide your physician in choosing a personalized treatment. However, testing is expensive and time-consuming, but not always available in major hospitals.
To explore faster and more accessible options, researchers trained AI with H&E staining pathology slides. This is what standard pink and purple tissue imaging pathologists use to diagnose cancer under the microscope. These slides are prepared from tumor samples collected with standard diagnostic biopsies and are routine parts of diagnostic workup in almost every patient.
“We asked. Can AI train them to predict genetic mutations using standard pathological slides, which are part of work-ups in all patients?” says Campanella. “This could support faster treatment decisions without compromising the quality of care.”
The team has developed a new AI model that fine-tunes a large “basic” model for specific tasks to predict EGFR (epidermal growth factor receptor) mutations from these slides. EGFR is a protein on the cell surface that helps grow and divide.
Mutations in the EGFR gene may promote cancer growth, particularly in patients with lung adenocarcinoma. It is important to identify these mutations. Because tumors are highly responsive to targeted therapy, but only if detected. While confirmation still requires advanced genetic testing, researchers are investigating ways that AI can help flag cases faster and more efficiently, making better use of limited tumor samples and accelerating the path to treatment.
Real-time, behind the scenes “silent trials” – the first of pathology, AI analyzed live patient samples at the Memorial Sloan Kettering Cancer Center. Although the prediction of AI was not visible to clinicians, it showed that it could reliably detect EGFR mutations and reduce the need for potentially rapid genetic testing by more than 40%, the researchers say. To demonstrate generalizability, data from hospitals in the US and Europe were analyzed retrospectively.
“This study, including known biomarkers, shows how AI can thinkfully integrate AI into cancer diagnosis to support faster, smarter, and more personalized care,” said the associate professor at MD, PhD, Windreich, Artificial Intelligence and Human Health, and Artificial Intelligence and Human Health, Psychiatry, Genetics, and Neuricience and Neusience and Neusience and Neusience. “Flagging previous important mutations can help oncologists act quickly, but reduce the burden of sequencing labs in high-resource settings that run rapid testing. Actual promises can not only be efficient, but also discover new biomarkers from everyday pathology.
The team plans to continue data collection through silent trials, extending it to additional sites and lay the foundation for the regulatory approval process. In the long term, the researchers aim to broaden the system's capabilities to detect additional cancer biomarkers and assess the impact in a lower resource setting where access to genetic testing is more limited. Together, these efforts can lead to wider clinical adoption of AI and improve patient outcomes in both low- and high-resource settings.
The paper is titled “Enhancement of clinical genomics in lung adenocarcinoma with practical development of fine-tuned computational pathology basic models.”
The authors of the studies described in the journal include Gabriele Campanella, Neeraj Kumar, Swaraj Nanda, Siddharth Singh, Eugene Fulder, Ricky Kwan, Silke Muersted, Nicole PFAR, Peter J. Schferer, Ida Nator Mastean Alimurkel's 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, Chad Vanderbilt.
This work was supported by the AI-Reaide Mount Sinai (Air.MS) platform and the expertise of the Hasso Pratner Health Institute at Mount Sinai (HPI.MS). Computational resources and expertise are also utilized from the scientific computing and data of the ICAHN School of Medicine in Mount Sinai and supported by the Clinical and Translational Science Award (CTSA) Grant UL1TR004419.
Additionally, research funding was provided by the Warren Alpert Foundation through the Cancer Center Support Grant from NIH/NCI (P30CA008748) and the Warren Alpert Foundation through the Warren Alpert Center for Digital and Computational Pathology at the Memorial Sloan Kettering Cancer Center.
/Public release. This material of the Organization of Origin/Author is a point-in-time nature and may be edited for clarity, style and length. Mirage.news does not take any institutional position or aspect, and all views, positions and conclusions expressed here are the views of the authors alone.
