Prospective evaluation of third-generation small bowel capsule endoscopy videos by independent readers demonstrates poor reproducibility of cleanliness classifications.

OBJECTIVE: The detection of lesions during small bowel (SB) capsule endoscopy (CE) depends on the cleanliness of the intestine. Quality reporting and comparison of different preparation methods require reliable scores. Three scores known as quantitative index (QI), qualitative evaluation (QE), and overall adequacy assessment (OAA), have been proposed to assess SB cleanliness, and are sometimes used in clinical practice and in clinical trials. However, none of these scores has received any external validation. The aim of our study was to re-assess the reproducibility of these three specific scores. METHODS: One-hundred-and-fifty-five complete third-generation SB-CE video recordings were extracted from a multicenter randomized controlled trial (PREPINTEST) which evaluated three modalities of SB preparation for CE. Three experts independently read the 155 SB-CE video recordings twice, in a random order, over 48 -h periods at 6-week intervals, using the QI, QE and OAA scores. Cohen's linearly weighted kappa coefficients were calculated to assess intra-observer and inter-observer agreements. RESULTS: Intra-observer reproducibility was fair to moderate, with kappa coefficients between 0.37 and 0.46 for QI, 0.41 and 0.51 for QE, 0.41 and 0.50 for OAA. Inter-observer reproducibility was fair to substantial according to kappa coefficients between experts varying from 0.40 to 0.64, 0.29 to 0.65, and 0.52 to 0.71, for QI, QE and OAA, respectively. CONCLUSIONS: QI, QE and OAA scores, currently used for evaluation of the quality of the preparation of SB-CE, are not sufficiently reproducible. Other scores or methods are therefore needed for SB-CE cleanliness assessment.

A neural network-based algorithm for assessing the cleanliness of small bowel during capsule endoscopy.

BACKGROUND: Cleanliness scores in small-bowel capsule endoscopy (SBCE) have poor reproducibility. The aim of this study was to evaluate a neural network-based algorithm for automated assessment of small-bowel cleanliness during capsule endoscopy. METHODS: 600 normal third-generation SBCE still frames were categorized as "adequate" or "inadequate" in terms of cleanliness by three expert readers, according to a 10-point scale, and served as a training database. Then, 156 third-generation SBCE recordings were categorized in a consensual manner as "adequate" or "inadequate" in terms of cleanliness; this testing database was split into two independent 78-video subsets for the tuning and evaluation of the algorithm, respectively. RESULTS: Using a threshold of 79 % "adequate" still frames per video to achieve the best performance, the algorithm yielded a sensitivity of 90.3 %, specificity of 83.3 %, and accuracy of 89.7 %. The reproducibility was perfect. The mean calculation time per video was 3 (standard deviation 1) minutes. CONCLUSION: This neural network-based algorithm allowing automatic assessment of small-bowel cleanliness during capsule endoscopy was highly sensitive and paves the way for automated, standardized SBCE reports.

Multi-criterion, automated, high-performance, rapid tool for assessing mucosal visualization quality of still images in small bowel capsule endoscopy.

Background and study aims Capsule endoscopy (CE) is the preferred method for small bowel (SB) exploration. Its diagnostic yield can be reduced by poor mucosal visualization. We aimed to evaluate three electronic parameters - colorimetry, abundance of bubbles, and brightness - to assess the adequacy of mucosal visualization of SB-CE images. Patients and methods Six-hundred still images were randomly extracted from 30 complete and normal SB-CEs. Three experts independently evaluated these images according to a 10-point assessment grid. Any frame with a mean score above seven was considered adequately cleansed. Each image was analyzed electronically according to the three preset parameters, individually and then combined, with the experts' score as reference. A random forests methodology was used for machine learning and testing. Results The combination of the three electronic parameters achieved better discrimination of adequately from inadequately cleansed frames as compared to each individual parameter taken separately (sensitivity 90.0 % [95 %C. I. 84.1 - 95.9], specificity 87.7 % [95 %C. I. 81.3 - 94.2]). Conclusion This multi-criterion score constitutes a comprehensive, reproducible, reliable, automated and rapid cleansing score for SB-CE frames. A patent is pending at the European patent office.