Development and validation of a computed assessment of cleansing score for evaluation of quality of small-bowel visualization in capsule endoscopy.

BACKGROUND AND STUDY AIMS: An objective and reliable scoring system is needed to assess quality of visualization in small bowel (SB) capsule endoscopy (CE), for both clinical practice and research purposes. The aim of this study was to establish and to validate a SB-computed assessment of cleansing (SB-CAC) score. PATIENTS AND METHODS: Thirty-three SB-CE were selected. A CAC score, defined as the ratio of the red over green pixels (R/G ratio), was calculated for each frame. Intervals were then determined, ranging from the lowest to the highest ratio among the extracted frames. Twelve frames were randomly selected in each of these intervals. Two hundred eighty-eight frames were shuffled and analyzed twice in random order by two experienced CE readers who were blinded to the CAC scores. Once an "adequately cleansed" or "inadequately cleansed" qualification was allotted to every still frame, a receiver operating characteristic (ROC) curve was created. In case of discrepancy between the two readers, the still frames were excluded. A second dataset of 288 different SB still frames was generated and read twice in random order by two other experienced SB-CE readers, using the same methodology. RESULTS: A SB-CAC score threshold of 1.6 best achieved discrimination of adequately from inadequately cleansed frames, with a sensitivity of 92.7 % (95 %CI [89.7 - 95.8]) and a specificity of 92.9 % (95 %CI [89.9 - 95.9]). This threshold was validated using the second dataset, yielding the following performances: sensitivity 91.3 % (95 %CI [87.9 - 94.6]), specificity 94.7 % (95 %CI [92.1 - 97.3]). CONCLUSION: An SB-CAC score of 1.6 has the highest sensitivity and specificity to discriminate "adequately cleansed" from "inadequately cleansed" SB-CE still frames. This constitutes an objective, reproducible, reliable, and automated cleansing score for SB-CE.