American College of Physicians, Maryland Regional Associates Meeting, Baltimore Maryland
Accuracy of the suspected blood indicator in detecting red lesions during wireless capsule endoscopy
May 30, 2007
Background: Wireless capsule endoscopy is an innovative technology most commonly used in the evaluation of obscure gastrointestinal bleeding. As the capsule travels through the digestive system, it collects in excess of 50,000 images, all of which must be reviewed by a gastroenterologist. This can be time consuming, requiring as much as 90 minutes to review a single study. The "Suspected Blood Indicator" (SBI) is a software feature proposed to aid in the detection of lesions and speed reading times. The purpose of this study was to evaluate the ability of the SBI to accurately identify red lesions during wireless capsule endoscopy.
Methods: A retrospective analysis of 154 consecutive capsule studies was carried out on cases performed between July 20, 2004 and August 9, 2006. The PillCam™ SB Capsule Endoscope (Given Imaging, Yoqneam, Israel) was used in this study. Demographic data, medications, comorbid conditions, gastrointestinal imaging and endoscopic studies were reviewed. The principal outcome variables were actual lesions identified by a gastroenterologist and presumed lesions identified by the SBI.
Results: A total of 151 studies were included, with 3 excluded from the analysis due to a failure of the capsules to leave the stomach. Of these, there were 21 patients with active bleeding, 32 with patchy erythema suspected to be inflammation, 39 with non-bleeding arteriovenous malformations, 40 with ulcers/erosions, and 23 with tumors/polyps. The SBI had an overall sensitivity of 47.68% and specificity of 61.19%. For actively bleeding lesions, the overall sensitivity and specificity improved to 82.54% and 81.39%, respectively.
Conclusions: The sensitivity and specificity of the SBI are too low to be reliable as a screening or confirmatory tool, even for actively bleeding lesions. Therefore, it likely will not decrease the time needed to interpret a study, or improve the accuracy of those results. More advanced software techniques are needed, such as those which incorporate texture, location, probability theory, or artificial intelligence.