Advancing cancer care through artificial intelligence

In the rapidly evolving landscape of machine learning, the integration of artificial intelligence (AI) and cancer research and treatment is becoming increasingly vital. While physicians and cancer researchers have been using AI to aid in patient care for decades, it has played a pared-down role compared to where the field is today. 

Sylvia Plevritis, PhD, William M. Hume Professor in the School of Medicine, professor of biomedical data science and of radiology, and chair of the Department of Biomedical Data Science, was recently named the Stanford Cancer Institute associate director for cancer AI. In December 2024, her department was awarded $8.9 million from the Advanced Research Projects Agency for Health to develop AI-augmented support tools for cancer tumor boards. The first step in developing these tools is creating a multimodal oncology data lake, a virtual centralized repository that can hold and process large amounts of data.  

Part of Plevritis’s strategic vision to revolutionize cancer care, the multimodal oncology data will help create a comprehensive data asset that links various types of information related to cancer patients by integrating electronic health records, imaging data, pathology reports, and genomic information. Creating this data asset is Plevritis’s primary goal because the integration will provide a holistic view of each patient's journey through cancer care. By de-identifying (replacing personally identifiable information like names or addresses with a code) and storing this data in the cloud, the project aims to make this large data asset accessible for research and clinical applications across Stanford.

"The idea is to create a longitudinal patient history that incorporates all data modalities," Plevritis explained. "This will serve as a resource for everyone at Stanford, enabling researchers and clinicians to build AI-driven tools that can analyze and interpret complex datasets."

From patient matching to decision support

The first phase of the data lake project focuses on creating the "Find Patients Like Me" feature, which aims to assist physicians in identifying similar patients based on various data modalities. This tool will allow clinicians to access information about past tumor board discussions, treatment decisions, and outcomes for patients with similar profiles, ultimately enhancing their ability to make informed decisions.

"The goal is to ensure that physicians can find similar patients quickly and efficiently, and then to apply it in the context of the tumor board," she said. "By providing insights into past cases, we hope to support their decision-making process without overwhelming them with information."

A significant aspect of this initiative is building interdisciplinary teams that bring together Stanford data scientists, AI experts, clinical faculty, and translational researchers.  

"I’m excited because we’re building very interdisciplinary teams across our institution."

When dealing with such a wide breadth and depth of data, a collaborative approach is essential for developing robust AI tools that can effectively address the complexities of cancer care.

This initiative departs from traditional AI applications in cancer research. Historically, prediction models were built on cohort-specific data, often leading to limited generalizability across different patient populations. 

“When you do that, the numbers start decreasing. And so then you’re only working with hundreds of patients, and you’re building a model on a very limited data set,” Plevritis said. 

This can profoundly impact how physicians make decisions regarding patient care. One key advantage of the model being developed is its ability to analyze data from various points in a cancer patient's journey. 

"This model will give us a better understanding of the dynamic nature of cancer," she said. "By using data that hasn’t been utilized before, we can better predict future events based on a patient's complete history.”

While healthcare providers have traditionally observed cancer progression in individual patients, they haven't had a platform that integrates the entire continuum of care. This new approach will allow for a comprehensive patient experience analysis, enabling clinicians to learn from all available data.

Looking ahead, Plevritis is optimistic about the future of AI in cancer research and care. 

"In the next five years, I believe we’re going to make significant contributions with these tools," she asserted. 

While current systems rely on published literature for recommendations, the goal is to create a robust foundation model based on real patient data, thus paving the way for more effective and personalized cancer care.

By Kai Zheng

#CancerCare #AIandCancer #TumorBoard