Artificial intelligence (AI) continues to generate excitement across oncology, but its real-world clinical impact remains inconsistent. In this episode of Exploring AI in Oncology, Dr. Ashish Manne, a Gastrointestinal Medical Oncologist at The Ohio State University Comprehensive Cancer Center, Arthur G. James Cancer Hospital and Richard J. Solove Research Institute, shares his American Association for Cancer Research (AACR) presentation on AI-driven pathological subtyping in pancreatic cancer and highlights the critical shift that is moving AI from theoretical promise to practical application.

Bridging Innovation and Clinical Reality

Dr. Manne’s work spans translational research, clinical trials, and AI-driven innovation. While the first two are well-established in academic oncology, AI represents a new frontier. Rather than developing AI for its own sake, his team focuses on solving clinically meaningful problems—particularly in pancreatic cancer, where treatment options remain limited.

At the heart of his research is the use of digital pathology and AI to classify tumor subtypes. “We all know AI is promising in oncology, and digital pathology is booming,” he explained, “but the question for me is, how can I use it in my clinic?” By leveraging routinely available pathology slides, his approach avoids reliance on expensive genomic technologies that may not be accessible globally.

This accessibility is key. “Practicing in the United States, we focus on what we have, but there are other places throughout the world where there may not be RNA or DNA sequencing available,” he noted. AI models built on widely available data could help to close this gap and democratize precision oncology.

Subtyping Pancreatic Cancer

One of the most promising applications discussed is AI-driven classification of pancreatic cancer into classical and basal subtypes—categories with meaningful therapeutic implications. “These subtypes help us predict whether tumors are aggressive and whether they will respond to a particular chemotherapy,” Dr. Manne said. Classical tumors tend to respond better to 5-FU–based regimens like FOLFIRINOX, while basal subtypes are more aggressive and less responsive.

In practice, this could directly influence treatment decisions. “If I have a patient with a classical subtype, I might go with FOLFIRINOX,” he explained. “When I have a basal subtype, I might be more proactive, doing frequent scans or looking for clinical trials.” The shift from retrospective classification to proactive clinical strategy illustrates how AI can augment, rather than replace, physician judgment.

Why AI Isn’t Yet Driving Decisions

Despite rapid technological progress, AI has not yet become a core component of clinical decision making. According to Dr. Manne, trust and transparency remain major barriers. “We are at least five or six years away from using AI in making clinical decisions,” he said. A key issue is the “black box” nature of many models. “We don’t trust the models because they don’t give us methodology or reasoning. We are playing with patients’ lives.”

Currently, AI’s most tangible impact is in improving workflow efficiency. Tools like clinical documentation assistants and EMR summarization are already reducing administrative burden. “The best example is using AI to improve our workflow,” he noted.

Still, Dr. Manne sees an opportunity for clinicians to engage more actively. “We are not doing enough to experiment with AI in clinical aspects. There’s room for that. We can use AI like an assistant while we take the first steps.

Rethinking How AI Models Are Built

A recurring theme in the discussion was the need for specialization. Broad, one-size-fits-all AI models are unlikely to succeed in complex fields like oncology.

“For cancer, you need genomic data, pathology, and clinical context,” Dr. Manne explained. “You cannot take a model built for one specialty and apply it to another.” Instead, he advocates for a bottom-up approach: physician-led model development tailored to specific diseases, workflows, and care settings. “Rather than allowing AI companies to build a model and forcing us to use it, we should go the other way around.”

This approach could also improve adoption in community settings, where resources and workflows differ significantly from academic centers.

Key Takeaways

AI’s value in oncology lies in practical clinical application, not just innovation for its own sake.
Digital pathology-based models offer a scalable way to expand precision medicine globally.
Subtyping pancreatic cancer with AI can help guide treatment selection and clinical strategy.
Trust, transparency, and evidence remain major barriers to AI-driven decision-making.
Near-term impact of AI is strongest in workflow optimization rather than direct clinical decisions.
Physician-led, specialty-specific AI development may be the key to meaningful adoption.

As AI continues to evolve, its success in oncology may depend less on technological breakthroughs and more on thoughtful integration into the realities of patient care.

About the Speakers

Ashish Manne, MBBS, is an Associate Professor of Medical Oncology at The Ohio State University Comprehensive Cancer Center, Arthur G. James Cancer Hospital and Richard J. Solove Research Institute, where he is also Co-Director of the Pancreatic Cancer Multidisciplinary Clinic. He specializes in the treatment of pancreatic, colorectal, esophageal, liver, and biliary tract cancers. Dr. Manne’s research focuses on biomarker discovery, immunotherapy, and precision medicine to develop personalized treatments for pancreatic and hepatobiliary cancers.

Waqas Haque, MD, MPH, is a second-year Hematology/Oncology Fellow at the University of Chicago. He recently completed his Internal Medicine Residency at New York University (NYU) in a Clinical Investigator Track. Dr. Haque has balanced his patient care work with a variety of research projects. During his fellowship training at the University of Chicago, he plans to further his work in innovative clinical trial design, value-based care delivery to cancer patients, and clinical investigation.

AI-Driven Pathological Subtyping for Pancreatic Cancer: Moving From Promise to Practical Application