A multicenter study has shown that an open-source artificial intelligence model, SEE-AI, significantly improves lesion detection and reduces interpretation times for small bowel capsule endoscopy.
Researchers evaluated SEE-AI against conventional review methods using 249 capsule endoscopy examinations from six hospitals. The AI generated annotated videos highlighting lesion areas across eight categories.
Across 1,550 confirmed lesions, AI-assisted review achieved 98.8% per-lesion sensitivity, up from 86.4% with standard interpretation. Mean reading time dropped from 17.9 minutes to 13.7 minutes.
In patients with suspected small-bowel bleeding, per-lesion sensitivity for clinically relevant bleeding lesions improved from 82.8% to 98.2%, and reading time fell from 18.0 to 14.1 minutes.
SEE-AI remains fully open-source, supporting transparency and reproducibility. Researchers suggest AI-assisted capsule endoscopy could become the future standard of care for small bowel imaging. Further prospective studies are needed to confirm these results in broader clinical practice.