Deep Learning Algorithm Predicts Need for Crohn’s Disease Therapy

A study published last month in Therapeutic Advances in Gastroenterology demonstrated that a deep learning (DL) model can predict the need for biological therapy using capsule endoscopy (CE) videos of newly-diagnosed Crohn’s disease patients.

Mayo Clinic experts indicate that CE is useful in helping to determine the extent and severity of Crohn’s disease, along with monitoring a patient’s response to therapy, as the method is generally well-tolerated and less invasive than traditional endoscopy.

In the study, the researchers noted that predictors of disease progression and treatment response are lacking. CE can help combat this by providing clinicians with images of the entire digestive system, but each CE generates ten to twelve thousand images per patient for interpretation.

Artificial intelligence (AI), with its ability to rapidly analyze large amounts of data, has been proposed as a tool to help clinicians capture details picked up by CE videos that may help determine the best treatment course for a patient.

The research team stated that DL methods can accurately detect and grade inflammatory findings found in CE images from Crohn’s disease patients. However, the predictive capabilities of DL models to forecast Crohn’s disease outcomes have not been well-studied, leading the researchers to develop their own model.

They began by collecting CE data from 101 Crohn’s disease patients who underwent CE within six months of diagnosis. From there, they tasked the DL model with predicting each patient’s need for biological therapy using imaging data from their CEs.

These predictions were compared to a gastroenterologist’s analysis of the inflammatory index found in patient stool samples and the Lewis score of the CE videos.

The researchers found that the DL model achieved an accuracy of 81 percent and an area under the ROC curve (AUC) of 0.86 when tasked with predicting the need for biological therapy. The Lewis score achieved an AUC of 0.70.

Overall, the DL tool outperformed both the human gastroenterologist and the inflammatory index from patient samples.

These findings led the research team to conclude that DL-based analysis of CE imaging in newly diagnosed Crohn’s disease patients can serve as an accurate prediction method to determine which patients may need biological therapy.

However, they noted that the model must be validated in future studies.

"Predicting disease course and patient outcomes for Crohn’s Disease is one of the most critical clinical challenges in inflammatory bowel disease treatment. However, this research highlights the potential impact of AI on this process,” said Uri Kopylov, MD, director of IBD in the Department ofGastroenterology at Israel’s Sheba Medical Center and professor of medicine at Tel Aviv University, in a press release shared with HealthITAnalytics.

“By adopting AI in clinical practice, we can begin to use our wealth of knowledge and research in personalized medicine to drive improved patient outcomes and open the door to new possibilities for diagnosis and treatment,” he noted.