Post written by Leonardo Zorron Cheng Tao Pu, MD, MSc, from the Faculty of Health and Medical Sciences, University of Adelaide, Adelaide, South Australia, Australia, and the Department of Gastroenterology and Hepatology, Nagoya University Graduate School of Medicine, Nagoya, Aichi, Japan.
In this study, we have used artificial intelligence through deep learning to predict the histology of colorectal lesions. The prediction of histology was based on the 5 classes described in the Modified Sano’s (MS) classification (which includes a separate class for sessile-serrated adenomas/polyps as well as one for advanced non-invasive lesions). Once the computer-aided diagnosis (CAD) training was completed with NBI images from an Australian center, it was validated with NBI and BLI images from a partner center in Japan with similar results.
Although prediction of histology based on advanced imaging (eg, NBI with magnification) is important, accuracy greatly varies. CAD has demonstrated to be helpful in interpreting medical images and had great potential in aiding endoscopists to predict histology in real-time during colonoscopy.
This is the first study looking into a CAD for colorectal lesions able to differentiate serrated lesions and advanced polyps. In addition, we were able to show that using deep convolutional neural networks (a machine learning subset), a CAD is able to predict images on BLI even when solely trained on NBI. The next steps to integrate CAD into daily colonoscopy practice include the integration of detection and characterization modules, development of user-friendly interfaces, and multicenter randomized trials addressing the added value of CAD in real-world outcomes (eg, ADR and avoidance of surgery for non-invasive/superficial cancer).
Figure 3. Mean receiver operating characteristic (ROC) curves (AUC) and 95% confidence intervals per class for the exploratory phase (Australian dataset).
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.