Researchers use AI to predict rare cancer immunotherapy responses

Researchers from The University of Texas MD Anderson Cancer Center demonstrated that an artificial intelligence (AI)-based analysis of tumor biopsies can predict responses to immunotherapy in a study of patients with rare cancers, published in the Journal for ImmunoTherapy of Cancer.

Led by Aung Naing, M.D., professor of Investigational Cancer Therapeutics, this analysis builds on recently published research that identified features in the tumor microenvironment that were predictive of immunotherapy response in patients with rare cancers, even in those who did not have known markers of immunotherapy response.

AI-based pathology has the potential to provide clinicians with useful information on both the tumor and its surrounding microenvironment, helping to guide personalized treatment decisions for patients receiving immunotherapy." 

Aung Naing, M.D., Professor of Investigational Cancer Therapeutics

How does this AI tool work and what are its advantages in guiding immunotherapy treatment in rare cancers?

Naing's previous publication identified two features that could best indicate whether patients were responding to immunotherapy. These included how many immune cells were present within the tumor before treatment and changes in immune cell infiltration into the tumor during treatment.

Manually counting individual immune and cancer cells on pathology slides requires significant effort, especially when trying to scale the effort to large numbers of slides and patients, but AI tools can do this quickly. In the current study, the AI-based analysis rapidly generated these measurements and tracked changes longitudinally across multiple biopsies from the same patients. It is also notable that this approach utilizes standard pathology slides that are already routinely collected.

How did this approach perform and what is next for this research?

While both an increase in tumor immune infiltration and a decrease in tumor content were predictive metrics on their own, these individual signals were much stronger when combined. This pattern reflects both an active immune response and a reduction in tumor burden.

Patients with favorable signals had a 64% lower risk of disease progression or death and lived nearly four times longer on average (median survival of 42 months vs. 10 months) compared to patients without these markers.

While these results are promising, validation in larger patient populations is needed before this approach is ready to guide treatment decisions in the clinic.

"While this AI-powered approach needs validation, this is an exciting step forward because it shows that meaningful insights can be extracted from routine pathology samples across a diverse group of rare cancers," Naing said.