The following is a summary of “Deep learning based lesions detection and severity grading of small bowel Crohn’s disease ulcers on double-balloon endoscopy images,” published in the December 2023 issue of Gastroenterology by Xie et al.

Despite the prevalence of double-balloon endoscopy (DBE) for diagnosing Crohn’s disease (CD), misdiagnosis remains a risk due to inexperience and inherent subjectivity in lesion detection.

Researchers started a retrospective study to leverage artificial intelligence (AI) in detecting and objectively assessing small bowel CD, paving the way for more refined disease management.

They acquired 28,155 small bowel DBE images from 628 patients (January 2018 to December 2022). Four gastroenterology experts annotated the photos, and the final decision, with a consensus agreement from at least two endoscopists, was determined. Using EfficientNet-b5, the model was trained to detect various CD lesions, including ulcers, non-inflammatory stenosis, and inflammatory stenosis. The ulcer grading considered the ulcerated surface, size, and depth. An evaluation comparing the AI model’s performance with that of endoscopists was conducted.

The results showed EfficientNet-b5 achieved impressive accuracies, precisely 96.3% (95% CI, 95.7%-96.7%), 95.7% (95% CI, 95.1%-96.2%), and 96.7% (95% CI, 96.2%-97.2%) for detecting ulcers, non-inflammatory stenosis, and inflammatory stenosis, respectively. In the realm of ulcer grading, EfficientNet-b5 demonstrated solid performance with average accuracies of 87.3% (95% CI, 84.6%-89.6%) for grading the ulcerated surface, 87.8% (95% CI, 85.0%-90.2%) for grading the size of ulcers, and 85.2% (95% CI, 83.2%-87.0%) for assessing ulcer depth.

They concluded that the efficientNet-b5 AI model nailed CD lesion detection and grading, boosting objectivity and paving the way for optimal treatment.

Source: sciencedirect.com/science/article/abs/pii/S001651072303136X