Post written by Babu P. Mohan, MD, from the Gastroenterology & Hepatology, University of Utah, Salt Lake City, Utah, USA.

The study aimed to analyze the pooled rates of diagnostic accuracy parameters of computer-aided diagnosis (CAD) by convolutional neural networks (CNN)-based machine learning of wireless capsule endoscopy (WCE) images.

Recent studies have reported on the use of machine learning algorithms in CAD of GI ulcers and/or hemorrhage on WCE study. Studies used multiple different types of machine learning algorithms; however, CNN is a type of deep learning algorithm that has demonstrated outstanding performance in identification of objects by image analysis. We did this study to exclusively assess the diagnostic accuracy of CNN algorithms in WCE image analysis.

As a result of our exhaustive literature search, we found 9 studies that were included in our final analysis. CNN-based CAD of WCE images demonstrated high pooled accuracy parameters (accuracy=95.4%, sensitivity=95.5%, specificity=95.8%, positive predictive value=95.8%, and negative predictive value=96.8%). Heterogeneity was negligible. Based on our study results, future studies should focus on machine learning algorithms that solely utilize CNN-based models. Real life studies are warranted to evaluate how a CNN-based CAD would perform and help the physician endoscopist in WCE reading and reporting. We hypothesize that a CNN-based CAD tool in WCE image analysis will help in automated reading and reporting of lesions in an incredibly short amount of time.  

Read the full article online.

The information presented in Endoscopedia reflects the opinions of the authors and does not represent the position of the American Society for Gastrointestinal Endoscopy (ASGE). ASGE expressly disclaims any warranties or guarantees, expressed or implied, and is not liable for damages of any kind in connection with the material, information, or procedures set forth.