AI-Assisted Double-Headed Capsule Endoscopy: Multicentre Prospective Diagnostic Accuracy Study Across Small Bowel Indications

Abstract

Background/Aims: Double-headed capsule endoscopy enhances visualization and diagnostic yield in small bowel evaluation but increases reading time. This study aimed to assess the diagnostic performance of AI-assisted double-headed capsule endoscopy (MiroCam MC2000) across all small bowel indications and to compare its reading efficiency with the standard manual reading mode. Methods: From May to December 2023, 242 consecutive patients (mean age 50.17 years, SD 18.3; 53% female) underwent small bowel capsule endoscopy at two UK centres for suspected Crohn's disease (48.8%), iron-deficiency anemia (23.6%), bleeding (18.6%), or other (9%). Seven experienced readers reviewed videos in standard mode (blinded to clinical data), then AI-assisted (MiroCam AI Scan) methods were applied after de-identification/randomization. Two experts provided reference standards. No adverse events occurred. Results: AI-assisted reading had sensitivity 95.3% (95% CI 90.1-98.3%) and specificity 96.5% (95% CI 91.3-99.0%) for diagnostic findings, vs. standard reading: 96.5% (95% CI 91.2-99.0%) and 85.3% (95% CI 78.0-90.9%). The positive findings rate was 83.6% vs. 80.2% (p = 0.040). Reading time decreased by 52% (38.1 vs. 18.26 min; p < 0.001). Conclusions: AI-assisted reading offers high diagnostic accuracy, superior specificity and reduced reading times, supporting its adjunctive role with expert oversight. Registered: ERGO ID 82419.