Improved Efficiency and Lesion Detection in Small Bowel Capsule Endoscopy Using the Open-Source Artificial Intelligence Model SEE-AI.
SEE-AI significantly improved lesion detection rates and reduced reading time in small bowel capsule endoscopy compared to conventional methods.
Where it sits
this study against the rest of the glutathione (gsh) corpusSummary and findings
This study evaluated the effectiveness of an open-source AI model, SEE-AI, in improving lesion detection during small bowel capsule endoscopy (CE) compared to conventional reading. The analysis included 249 examinations across six hospitals, focusing on sensitivity and reading time. Results indicated significant improvements in both detection rates and efficiency.
Abstract
<h4>Objectives</h4>Small bowel capsule endoscopy (CE) produces lengthy videos that are time-consuming to review and susceptible to missed lesions. We evaluated whether an open-source, pretrained artificial intelligence (AI) model (SEE-AI) could improve diagnostic performance and interpretation efficiency compared with conventional reading.<h4>Methods</h4>We retrospectively analyzed 249 PillCam SB3 examinations performed between 2007 and 2022 at six hospitals, using a two-reader crossover design. SEE-AI (confidence threshold 0.1) generated annotated videos with bounding boxes for eight lesion categories. The primary endpoints were sensitivity for lesion detection on a per-lesion and per-patient basis. Secondary endpoints included specificity, predictive values, overall accuracy, and reading time. A prespecified subgroup analysis evaluated cases of suspected small-bowel bleeding (SSBB), focusing on Saurin P1+P2 hemorrhagic lesions.<h4>Results</h4>Across 1550 adjudicated lesions, AI-assisted reading demonstrated higher sensitivity than conventional reading (per-lesion: 98.8% [1532/1550] vs. 86.4% [1339/1550]; per-patient: 99.1% [464/468] vs. 80.3% [376/468]; both <i>p</i> < 0.0001). The mean reading time decreased from 17.9 to 13.7 min (<i>p</i> < 0.0001). In SSBB cases (<i>n</i> = 131), sensitivity for P1+P2 lesions improved on both a per-lesion basis (98.2% [439/447] vs. 82.8% [370/447]) and per-patient basis (98.6% [145/147] vs. 73.5% [108/147]), with a shorter reading time (14.1 vs. 18.0 min; all <i>p</i> < 0.0001).<h4>Conclusions</h4>In this multicenter evaluation, SEE-AI significantly improved lesion detection and reduced reading time for CE interpretation, including SSBB cases, while maintaining openness and reproducibility. AI-assisted reading may reduce clinicians' workload and support the adoption of SEE-AI as a practical tool - and a potential future standard of care - for small bowel CE.<h4>Trial registration</h4>N/A.
Background
This paper addresses the challenge of missed lesions during small bowel capsule endoscopy (CE), which can lead to significant diagnostic delays. Prior studies have indicated that AI can enhance diagnostic accuracy, but the effectiveness of specific models like SEE-AI had not been thoroughly evaluated in a multicenter setting. This study aims to assess whether SEE-AI can improve both the sensitivity of lesion detection and the efficiency of the reading process.
Methods
The study employed a retrospective analysis of 249 PillCam SB3 examinations conducted between 2007 and 2022 across six hospitals. A two-reader crossover design was utilized, with SEE-AI generating annotated videos for eight lesion categories. Primary endpoints included sensitivity for lesion detection on both per-lesion and per-patient bases, while secondary endpoints encompassed specificity, predictive values, overall accuracy, and reading time.
Results
The primary endpoint revealed that AI-assisted reading achieved a per-lesion sensitivity of 98.8% (1532/1550) compared to 86.4% (1339/1550) for conventional reading, with a p-value of <0.0001. Additionally, per-patient sensitivity was 99.1% (464/468) for AI versus 80.3% (376/468) for conventional methods, also with a p-value of <0.0001. The mean reading time was significantly reduced from 17.9 minutes to 13.7 minutes (p<0.0001). In cases of suspected small-bowel bleeding, sensitivity for P1+P2 lesions improved to 98.2% (439/447) for AI compared to 82.8% (370/447) for conventional reading, p<0.0001.
Interpretation
The findings suggest that SEE-AI significantly enhances lesion detection rates compared to traditional methods, which aligns with previous literature supporting AI's role in diagnostic imaging. However, while the statistical significance is clear, the clinical relevance of the sensitivity improvements should be considered in the context of potential confounding factors such as the retrospective design and the relatively small sample size in the subgroup analysis. These results imply that SEE-AI could be a valuable tool in clinical practice, but further prospective studies are warranted to confirm these findings.
Key findings
- Per-lesion sensitivity: 98.8% (1532/1550) for AI vs 86.4% (1339/1550) for conventional reading, p<0.0001.
- Per-patient sensitivity: 99.1% (464/468) for AI vs 80.3% (376/468) for conventional reading, p<0.0001.
- Mean reading time decreased from 17.9 to 13.7 minutes, p<0.0001.
- In suspected small-bowel bleeding cases (n=131), per-lesion sensitivity for P1+P2 lesions: 98.2% (439/447) for AI vs 82.8% (370/447), p<0.0001.
- In suspected small-bowel bleeding cases, per-patient sensitivity: 98.6% (145/147) for AI vs 73.5% (108/147), p<0.0001.
- Mean reading time in suspected small-bowel bleeding cases: 14.1 minutes for AI vs 18.0 minutes for conventional reading, p<0.0001.
Limitations
- Retrospective analysis may introduce bias.
- Small sample size in subgroup analysis for suspected small-bowel bleeding.
- Single-site data collection may limit generalizability.