NCI-funded researchers combined long-term, patient-outcome data with pathology slides from people with colorectal cancer (CRC) to develop a machine learning tool, called QuantCRC. Using this tool, the researchers could predict whether a patient’s cancer would recur based on analysis of a single hematoxylin and eosin (H&E) stained slide of the tumor.

CRC varies considerably from patient to patient. Differences in the tumor’s molecular characteristics, cellular environment, as well as the individual patient’s immune response, make it difficult for pathologists and oncologists to predict tumor behavior. Knowing which patients will respond to treatment and if the cancer will recur would be a significant advance in precision medicine, giving pathologists a data-driven tool for understanding CRC’s molecular underpinnings. 

The investigators applied QuantCRC to 6,468 digitized CRC H&E slides and extracted 15 features from each image. Those features captured details about the tumor and its environment (i.e., tumor-to-stroma ratio, tumor budding, tumor infiltrating lymphocytes, tumor grade, mucin, and tumor necrosis, among others).

The researchers found that QuantCRC performed on par with pathologists in interpreting tumor morphology. They also found the 15 features were strongly associated with the tumor’s molecular characteristics. Using these 15 features, they developed and tested a model to predict recurrence, giving oncologists another tool to help guide therapy and follow-up in patients with CRC.

According to Dr. Rish K. Pai, corresponding author on this study, “QuantCRC lets us identify different regions within the tumor and extract quantitative data from these regions.” He added, “The large number of tumors that we analyzed helped us learn which features were most predictive of tumor behavior. Now, we can apply what we have learned to new colon cancers to predict how those tumors will behave.”

The next steps will be to test the tool prospectively, in a real-world clinical environment.

Dr. Danielle Carrick, Program Director in the Genomic Epidemiology Branch in NCI’s Division of Cancer Control and Population Sciences, noted, “QuantCRC, and other tools like it, could aid pathologists in their work characterizing tumors and may help provide better estimates of risk of tumor recurrence, particularly for morphologically heterogeneous tumors such as CRC.”