Research at Sheba Medical Center at Tel Hashomer has shown that machine-learning analysis of complete capsule endoscopy (CE) videos at initial diagnosis can accurately predict the need for biological therapy to treat the gastroenterological disorder known as Crohn’s disease (CD).
Algorithms analyzing such videos at initial diagnosis of CD achieved 81% accuracy, significantly higher than a gastroenterologist using doctoral analysis of the inflammatory index in stool samples (calprotectin).
CD, a chronic condition in which sections of the digestive system become inflamed, affects people of all ages and whose symptoms usually begin in childhood or early adulthood. These include diarrhea, fatigue, stomach aches, cramps, weight loss, and blood in the stool.
Led by Prof. Uri Kopylov, Director Of Inflammatory Bowel Disease Service at Sheba; Prof. Shomron Ben-Horin, director of the gastroenterology department; and Intel engineering director Amit Bleiweiss, the team published their paper in the Therapeutic Advances in Gastroenterology journal under the title “Spatiotemporal analysis of small bowel capsule endoscopy videos for outcomes prediction in Crohn’s disease.”
How does artificial intelligence help treat Crohn's disease
The Sheba doctors and data researchers trialed a newly developed deep-learning model on complete CE videos from 101 CD patients, achieving an 81% accuracy level.
“Predicting disease course and patient outcomes for the disease is one of the most critical clinical challenges in inflammatory bowel disease treatment, but our research highlights the potential impact of AI on this process,” said Kopylov. “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.”
In CD, predictors of disease prognosis and response to treatment are still unknown. Capsule endoscopy allows for the analysis of the entire digestive system using a microscopic device equipped with a transmitter and camera, but every capsule film produced includes 10,000 to 12.000 images for interpretation. Due to the large amount of visual information in each video, it’s hard for a doctor to discern all necessary details – but these can be picked up by AI algorithms instead.
This research followed a trial last year in which the AI algorithm demonstrated it could scan a film of up to 12,000 images in two minutes. The research also found AI to be a highly effective diagnostic tool, producing 86% accuracy in image and data analysis compared to 68% accuracy achieved with the reliance on analysis by an experienced gastroenterologist. AI analysis was also compared to analysis of the inflammatory index in stool (calprotectin) by an experienced physician.
“Our findings are further proof of the powerful impact that AI can have in transforming our health systems and driving positive patient outcomes,” concluded Dr. Eyal Klang, head of the Sami Sagol AI Hub at Sheba’s ARC Innovation Center. “Building on our successful collaboration with Intel, we are looking ahead to further validations of this technology and seeing it implemented in hospitals and clinics worldwide.”