AI Model Improves Colorectal Cancer Recurrence Prediction

According to a Mayo Clinic-led study published in Gastroenterology earlier this month, an artificial intelligence (AI) model that incorporates a deep-learning framework may improve predictions of recurrence and survival in colorectal cancer (CRC) patients.

According to the Centers for Disease Control & Prevention (CDC), CRC is the fourth most common cancer in men and women and the fourth leading cause of cancer-related deaths in the US. The American Cancer Society estimates that there will be 106,180 new cases of colon cancer and 44,850 new cases of rectal cancer in the US this year.

Accurate cancer recurrence predictions for CRC patients are critical to improving patient outcomes and survival rates. However, predicting recurrence and survival relies on multiple factors and can be difficult. To address these challenges, the researchers sought to develop an AI model that could predict cancer recurrence using tumor images.

The research team began by gathering 6,500 digital slide images of CRC tumors. These images were used to develop a deep-learning algorithm, known as QuantCRC, that could identify different regions of interest within the tumors. From these regions, 15 prediction parameters were pulled from each image and compared with the findings noted in the corresponding patient’s EHR and pathology reports to verify QuantCRC’s ability to identify clinically relevant image features.

Then, the researchers build a prognostic AI model using QuantCRC to predict recurrence-free survival. To train the model, the research team utilized CRC biospecimens from participating Australian, Canadian, and US locations in the Colon Cancer Family Registry. The model was validated using data from an external cohort of locations in the US and Canada not participating in the registry.

Overall, the model achieved high predictive performance. The incorporation of QuantCRC significantly improved the predictions of the prognostic model. According to the press release, the model can identify subsets of patients who may or may not need chemotherapy or other intensive treatments based on their probability of cancer recurrence.

"QuantCRC can identify different regions within the tumor and extract quantitative data from these regions," said Rish Pai, MD, PhD, a pathologist at Mayo Clinic and senior author of the study, in the press release. "The algorithm converts an image into a set of numbers that is unique to that tumor. The large number of tumors that we analyzed allowed us to learn which features were most predictive of tumor behavior. We can now apply what we have learned to new colon cancers to predict how the tumor will behave."

This research is the latest in a string of efforts to use AI to enhance cancer care, medical imaging, and chronic disease management.

 Last year, researchers found that an AI model can detect and diagnose CRC on par with or better than pathologists by analyzing tissue scans.

In April, research revealed that using an AI-based colonoscopy screening tool may potentially prevent CRC incidence, reduce mortality, and increase cost-effectiveness across the care continuum.