Automatic detection of blood content in capsule endoscopy images based on a deep convolutional neural network.

BACKGROUND AND AIM: Detecting blood content in the gastrointestinal tract is one of the crucial applications of capsule endoscopy (CE). The suspected blood indicator (SBI) is a conventional tool used to automatically tag images depicting possible bleeding in the reading system. We aim to develop a deep learning-based system to detect blood content in images and compare its performance with that of the SBI. METHODS: We trained a deep convolutional neural network (CNN) system, using 27 847 CE images (6503 images depicting blood content from 29 patients and 21 344 images of normal mucosa from 12 patients). We assessed its performance by calculating the area under the receiver operating characteristic curve (ROC-AUC) and its sensitivity, specificity, and accuracy, using an independent test set of 10 208 small-bowel images (208 images depicting blood content and 10 000 images of normal mucosa). The performance of the CNN was compared with that of the SBI, in individual image analysis, using the same test set. RESULTS: The AUC for the detection of blood content was 0.9998. The sensitivity, specificity, and accuracy of the CNN were 96.63%, 99.96%, and 99.89%, respectively, at a cut-off value of 0.5 for the probability score, which were significantly higher than those of the SBI (76.92%, 99.82%, and 99.35%, respectively). The trained CNN required 250 s to evaluate 10 208 test images. CONCLUSIONS: We developed and tested the CNN-based detection system for blood content in CE images. This system has the potential to outperform the SBI system, and the patient-level analyses on larger studies are required.

The capsule endoscopy "suspected blood indicator" (SBI) for detection of active small bowel bleeding: no active bleeding in case of negative SBI.

OBJECTIVE: Capsule endoscopy (CE) is the gold standard to diagnose small bowel bleeding. The "suspected blood indicator" (SBI) offers an automated detection of active small bowel bleeding but validity of this technique is unknown. The objective was to analyze specificity and sensitivity of the SBI using the second small bowel capsule generation for the detection of active bleeding. METHODS: This is a retrospective analysis of all patients (199) who attended our clinic for CE from June 2008 through March 2013. The second-generation PillCam SB 2 capsule was used for detection of (1) luminal blood content and (2) potentially responsible small bowel lesions. The findings of an independent investigator were correlated to SBI findings and a number of SBI markings were analyzed by a receiver operating characteristic (ROC). RESULTS: In 157/199 cases, no sign of active bleeding or altered blood was detected. One hundred and thirty-seven of these 157 cases provided at least one SBI marking and a mean of 18.4 positive SBI markings per record were found. In 20 cases, neither SBI nor the human investigator detected abnormalities. Thirteen patients showed investigator-detected minor bleeding with mean SBI findings of 36 positive screenshots per record. When major bleeding was diagnosed by the investigator (n = 29), SBI detected a mean of 46.6 SBI-positive markings. SBI turned positive in 179 patients, whereas the investigator detected active bleeding in 42 cases. All patients with active bleeding were detected by SBI (sensitivity 100%, specificity 13%). ROC analysis revealed 51.0 SBI markings being the optimal cutoff for active versus no bleeding (sensitivity 79.1%, specificity 90.4%, misclassification of 15.3%). CONCLUSION: The new SBI software is a reliable tool to exclude active bleeding and/or major lesions but analysis of the CE video by a trained investigator is still important for the detection of lesions responsible for past bleeding.

The validity of suspected blood indicator software in capsule endoscopy: a systematic review and meta-analysis.

INTRODUCTION: Software tools have been developed as capsule endoscopy (CE) reading aids. The suspected blood indicator (SBI) tags possible areas of haemorrhage in the gastrointestinal (GI) tract. This meta-analysis aims to investigate the diagnostic accuracy of SBI in CE. Areas covered: A systematic literature search was conducted for studies on the use of SBI in CE. Sensitivity, specificity and diagnostic odds ratio (DOR) of SBI in diagnosing GI bleeding was evaluated. 2040 patients from 16 studies underwent 2049 CE examinations. The overall sensitivity of SBI for bleeding or potentially bleeding lesions was 0.553, specificity 0.578, DOR 12.354. The sensitivity of SBI for active bleeding was 0.988, specificity 0.646, DOR 229.89. Expert commentary: Currently, SBI has limited validity in CE reading. However, in active GI bleeding, it has good sensitivity, supporting its use in the acute setting.

Sensitivity of the suspected blood indicator: an experimental study.

AIM: To investigate whether suspected blood indicator (SBI) in capsule endoscopy (CE) is affected by background color and capsule passage velocity. METHODS: Experimental models of the small intestine constructed from paper in a variety of colors were used to simulate the background colors observed in CE images. The background colors studied included very pale yellow, yellow, very pale magenta, light grayish pink, burnt sienna, and deep and dark brown, and red spots were attached inside them. An endoscopic capsule was manually passed through the models. The rate of detection of the red spots by the SBI was evaluated based on the colors of the models and the capsule passage velocities (0.5 cm/s, 1 cm/s, and 2 cm/s). RESULTS: The rate of detection of the red spots by the SBI differed significantly according to the background color of the model (P < 0.001). Detection rates were highest for backgrounds of very pale magenta, burnt sienna, and yellow, in that order. They were lowest for backgrounds of dark brown and very pale yellow. The rate of detection of red spots by the SBI tended to decrease at rapid capsule passage velocities (1-2 cm/s) compared to slow velocities (0.5 cm/s) for backgrounds of very pale yellow (P = 0.042), yellow (P = 0.001), very pale magenta (P = 0.002), and burnt sienna (P = 0.001). No significant differences in the rate of detection were observed according to velocity for light grayish pink (P = 0.643) or dark brown (P = 0.396). CONCLUSION: SBI sensitivity was affected by background color and capsule passage velocity in the models. These findings may facilitate the rapid detection of bleeding lesions by CE.