Post written by Neal Mahajan, ScB, from the Division of Gastroenterology, Beth Israel Deaconess Medical Center, Boston, Massachusetts, and Indiana University School of Medicine, Indianapolis, Indiana, USA.

Our team has worked for several years on the use of machine learning (ML) in endoscopy and helped validate its additive effect in the endoscopy suite. We have noticed that there is very strong interest among gastroenterologists to learn how artificial intelligence works and how to use it, but there is a high barrier to entry because of the complexity of state-of-the-art ML models and the rapid rate of innovation in the space.

We identified automated ML (AutoML) as an important tool that can introduce ML to physicians without requiring much coding knowledge or understanding of current ML techniques. Our article demonstrates how the Google AutoML platform (Google, Mountain View, Calif, USA) can be used to create computer-aided detection and computer-aided diagnosis models based on the publicly available SUN Dataset.

We provide a tutorial for how to use the Google AutoML platform and the scripts we used to generate our annotation files, so that interested endoscopists can learn step by step how to use this technology and hopefully apply it to any novel endoscopic problems where they think computer vision could be helpful.

ML is being explored for a rapidly growing array of tasks relevant to endoscopy, but most endoscopists have yet to receive training on the basics of how it works. This project can be a useful starting point for endoscopists to learn how ML works and its strengths and weaknesses. This will allow more endoscopists to innovate using ML in a thoughtful and responsible manner.

Endoscopists can learn that getting into ML doesn’t have to be daunting! Although diving into the deep end of the technical aspects of modern ML and computer vision algorithms is a difficult task for anyone, tools such as AutoML can reduce the responsibility on you, allow you to focus on the high-level concepts of your ML project, and enable the platform to handle the technical challenges.

This is especially useful because keeping up with the constant improvements in ML is an uphill battle, and you can instead rely on AutoML to keep pace with these improvements for you.

Confusion matrix for Google AutoML computer-aided diagnosis model.

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.