Kidney cancer: Microscopic image of clear cell carcinoma, the most common type of renal cell carcinoma characterized by cytoplasmic clearing and a pattern of small branching blood vessels.
Kidney cancer: Microscopic image of clear cell carcinoma, the most common type of renal cell carcinoma characterized by cytoplasmic clearing and a pattern of small branching blood vessels. [Md Saiful Islam Khan/Getty Images]

Investigators at Dana-Farber Cancer Institute (DFCI), reporting in Cell Reports Medicine, say that a deep learning model they have developed can identify previously underappreciated features in clear cell renal carcinoma (ccRCC) from a two-dimensional pathology slide including tumor microheterogeneity that could help predict this rare kidney cancer’s response to immunotherapy.

According to the researchers, their findings show the pathology slides of ccRCC tumors contain important biological information that could be useful to better understand the biology of this form of cancer. Further, they noted, their results could also have the potential to be used for other types of cancer. This new study is one component of a broader effort at DFCI to leverage AI tools to gain a better understanding of tumor biology and find ways to improve cancer care.

“This is an example of the growing convergence of AI and cancer biology,” said Eliezer Van Allen, MD, chief of the division of population sciences at Dana-Farber and co-senior author of the paper. “It represents a major opportunity to measure key features of the tumor and its immune microenvironment at the same time. These measures could help drive not only biological discovery but also potentially guide cancer care.”

Renal cell carcinoma is one of the 10 most common cancers in the world, and ccRCC accounts for the vast majority—between 75% and 80%—of metastatic cases. While some tumors have responded to immune checkpoint inhibitors (ICIs), to date there is no diagnostic measure that can be used to determine whether or not a specific tumor will respond to ICIs.

According to Jackson Nyman, PhD, a graduate student in the Van Allen Lab at DFCI, the AI tool was first trained to assess a tumor’s nuclear grade. The nuclear grade of a tumor is assessed using stained tumor samples on a slide that describe how far tumor cells deviate from normal cells. The model the team developed not only successfully assessed nuclear grade, but showed it also can identify differences in grade across the tumor sample.

“We wanted to know what a tumor that responds to immunotherapy looks like. Is there anything in the pathology slide that might give us clues about what is different about the tumors?” noted Nyman who has since left DFCI for digital pathology company PathAI.

Their findings of the AI tool’s ability to identify nuclear grade inspired the team to further refine their model to quantify tumor microheterogeneity and immune properties, such as immune infiltration, across the slide. Microheterogeneity is the measure of how much nuclear grade variation is present in cells across the slide, while immune infiltration measures how deeply lymphocytes have penetrated the tumor. While pathologists have the ability to make these determinations on their own across a slide, they are very time-consuming making it impractical as part of their normal workflow.

When assessing the data from the AI model of the pathology slides, the researchers found that some tumors were notably homogenous, while others showed a range for different nuclear grades in different patterns. Likewise, some tumors showed the presence of lymphocytes while others lacked significant infiltration.

These differences perhaps hinted at whether certain patterns might be found in the tumor samples that could predict response to ICIs, Nyman noted.

To answer this question, the researchers turned to data from the CheckMate 025 randomized Phase III clinical trial which had tested monotherapy with an ICI or an mTOR inhibitor in patients with ccRCC who had been previously treated with standard therapy.

The AI tool revealed that features such as tumor microheterogeneity and immune infiltration were associated with improved overall survival among patients taking immune checkpoint inhibitors. The tumors that responded to ICIs had both higher levels of tumor microheterogeneity and denser infiltration of lymphocytes in high-grade regions.

“These signals are hiding in plain sight,” said Van Allen. “They are just hard for pathologists to practically measure on individual slides. With AI, we have a scalable way to potentially squeeze a lot more information out of these slides.”

The investigators are now testing the AI tool in a clinical trial involving a combination immunotherapy as a first-line treatment for ccRCC as it aims to gain its approval for clinical use.

Also of Interest