The study focused on the diagnostic accuracy of the TOP100 AI tool in Pillcam SB3 Capsule Endoscopy for patients with overt suspected small bowel bleeding. It involved a retrospective analysis of consecutive SBCE explorations from March 2018 to March 2021, adhering to strict inclusion and exclusion criteria. The primary endpoint was to assess TOP100's diagnostic performance for P2 lesions, with secondary endpoints including comparison of reading times between standard reading (SR) and TOP100 reading (TR), and the detection of predictors of concordance between TR and SR.
Results showed that TR detected 90.5% of patients with significant P2 lesions identified by SR, with a diagnostic accuracy of 92.79%. The AI tool exhibited high specificity (100%) and positive predictive value (100%) for P2 lesions, and even higher sensitivity (97%-100%) for active bleeding and angiodysplasias. However, its performance was inferior for P1 lesions, indicating the necessity of a second reading for these cases.
In terms of reading time, TOP100 significantly reduced the median reading time from 23 minutes for SR to 1.8 minutes for TR, marking a 91.7% reduction. This rapidity is crucial for the timely identification of bleeding sources and the scheduling of appropriate treatments.
The study also highlighted the importance of adequate intestinal cleansing for improving lesion detection with TOP100. Despite its high diagnostic accuracy, the study acknowledged the need for a second standard reading to fully exclude the presence of lesions, suggesting that the clinical impact of the miss rate should be evaluated in future prospective clinical trials.
Overall, TOP100 has proven to be an effective tool for the rapid review of SBCE videos, offering high sensitivity and specificity for detecting significant lesions, thereby facilitating early therapeutic decisions in patients with overt SBB.
