AI-Assisted Imaging to Assist Endoscopists in Colonoscopy Procedures

While AI tools for adenoma detection already exist, they have not significantly improved the identification of smaller or more subtle polyps. These tools mostly identify the adenomas that are already visible to most endoscopists. Researchers from the University of Texas at Austin (Austin, TX, USA) are focusing on detecting the more challenging, smaller polyps by analyzing each colonoscopy image pixel by pixel. Their AI algorithm is being trained to locate adenomas and precisely outline them, ensuring the tumor is detected and completely removed. The pixel-based segmentation helps address variations in colon contours, lighting issues, and even incomplete colon preparations during procedures. This advanced AI is designed to assist endoscopists in distinguishing between healthy and diseased tissue, even when the differences are subtle.

Currently, the only way to evaluate a polyp is through an endoscope, which provides a limited front view. However, the researchers are also developing a device that would allow endoscopists to “feel” the colon lining or a potential polyp before deciding whether to remove it. This device, an inflatable tactile sensor, would be attached to the endoscope. Once a polyp is identified, the endoscopist could move a donut-shaped balloon, about two inches in diameter, over the suspicious area. Using a joystick, the balloon would be inflated with air or liquid, and a sensor covering the balloon would provide tactile feedback about whether the area feels hard or soft, rough or smooth. (Hard or rough surfaces typically indicate a more problematic area.) This tactile sensor functions like a virtual fingertip on any surface and generates its own image, known as a vision-based tactile sensor. The team aims to have the tactile sensor ready for clinical trials within five years.

Related Links:
University of Texas at Austin