Post written by David R. Cave, MD, PhD, from Brigham and Women’s Hospital, Boston, Massachusetts, USA.

Video capsule endoscopy was first introduced in 2001. A major inhibitor to its widespread use has been that reading the output video takes significant time. The advent of artificial intelligence (AI) provides an opportunity to help the reader decrease time for reading a study.

The first generation of AI has demonstrated great potential, but an enormous amount of time and effort are needed for providing the AI system with training and validation. A second generation using multiple deep learning convolutional neural networks offers elimination of bounding of a training set and enhanced accuracy.

We have developed a method that sequentially uses 17 different convolutional neural networks to evaluate video capsule images (Figure 4). We have used a publicly available database as a model to demonstrate that we do not need to bound specific lesions, and we can enhance accuracy in the detection of mucosal abnormalities in that database.

We have shown that our methodology has achieved what we set out to do. The overall diagnostic accuracy was 99.79% with an accuracy of 100% for the detection of blood and foreign bodies. This was performed without bounding of lesions, thereby demonstrating that this provides a much more rapid means of developing AI for video capsules as they evolve with improved capabilities.

Furthermore, it will be possible to detect a much broader range of pathology than has been practical, even with limited data sets.

We have been able to run this technology on a laptop with a good-quality graphics card. We anticipate that when imaging devices are changed or upgraded this will make development of an appropriate AI system much quicker and less labor-intensive.

Classification of video capsule endoscopy images using a multiple convolutional neural network (CNN) approach. A high overall classification accuracy has been achieved.

Visit iGIE’s Facebook, Twitter, and YouTube accounts for more content from the ASGE peer-reviewed journal that launched in December 2022.

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.