AI-Assisted Capsule Endoscopy – First Multicentric Transatlantic Detection and Differentiation Validation Study

Aims Capsule endoscopy (CE) is a minimally invasive exam developed for small bowel evaluation, useful for detection of pleomorphic lesions. Nevertheless, its complex video analysis leads to lengthy reading times (30-120 minutes), with a risk of missing clinically relevant lesions. While artificial intelligence (AI) models have shown promise in enhancing diagnostic accuracy while reducing exam reading time, there is a lack of clinical validation of these preliminary results. In this context, the authors aimed to develop a multicentric prospective validation study comparing AI-assisted with conventional CE reading [1] [2] [3].

Methods The study included 137 CE videos from two different devices across four centers from both European and American continent. Following initial conventional reading reports, AI-assisted reading was performed by an independent expert using a deep learning model to detect and differentiate small bowel lesions (lymphangiectasias, xanthelasmas, vascular lesions, protruding lesions, ulcers, erosions and hematic residues). Both reports were reviewed by an expert from an independent center, which decided in discrepant cases, establishing the gold standard for CE analysis.

Results AI-assisted reading demonstrated superior performance compared to conventional reading. Sensitivity improved significantly (96.2% vs. 37.8%), as did specificity (99.0% vs. 87.3%) and overall accuracy (90.8% vs. 75.2%). AI-assisted reading detected 257 of 267 lesions identified in expert reviews, compared to only 101 by conventional reading. For specific lesion types, AI-assisted reading had higher diagnostic accuracy in the majority of categories: 87.6% vs. 38.7% for xanthelasmas and lymphangiectasias, 94.1% vs. 68.6% for vascular lesions, 86.1% vs. 87.6% for protuberant lesions, 86.2% vs. 83.2% for ulcers and erosions, and 100.0% vs. 97.8% for hematic residues. Additionally, AI-assisted reading was associated with a mean exam reading time to 239 seconds per exam.

Conclusions This was the first worldwide multicentric study proving that AI-assisted CE reading superiority compared to conventional reading, not only for lesion detection but also differentiation. Additionally, the validation study adressed the interoperability challenging by including videos from multiple devices. AI-assisted reading provides clear benefits for CE video analysis, advancing the current state of the art.

Publikationsverlauf

Artikel online veröffentlicht:
27. März 2025

• References

• 1 Afonso J.. et al. Automated detection of ulcers and erosions in capsule endoscopy images using a convolutional neural network. Med Biol Eng Comput 2022; 60: 719-725
• 2 Mascarenhas Saraiva M.J.. et al. Deep learning and capsule endoscopy: automatic identification and differentiation of small bowel lesions with distinct haemorrhagic potential using a convolutional neural network BMJ Open Gastroenterol. 2021; 8 doi:10.1136/bmjgast-2021-000753.
• 3 Nam J.H.. et al. Development of a deep learning-based software for calculating cleansing score in small bowel capsule endoscopy. Sci Rep 2021; 11: 4417