AI in Colonoscopy: Spotting Flat Sessile Serrated Polyps

Gastroenterologists are increasingly adopting real-time artificial intelligence (AI) during colonoscopies to improve the detection and characterization of flat sessile serrated polyps (SSPs). This article explores how Computer-Aided Detection (CADe) and Computer-Aided Diagnosis (CADx) systems work in tandem to identify these notoriously elusive, pre-cancerous lesions. By analyzing live video feeds frame-by-frame, AI acts as a second set of eyes for the physician, flagging subtle mucosal abnormalities, predicting polyp histology on the spot, and ultimately reducing the risk of interval colorectal cancer.

Flat sessile serrated polyps are a major contributor to colorectal cancer, yet they are among the hardest lesions to find during a standard colonoscopy. Unlike traditional adenomas, which often protrude visibly into the colon lumen, SSPs lie flat against the intestinal wall. They typically have indistinct borders, a pale color that blends in with the surrounding healthy mucosa, and are frequently covered by a thin layer of mucus. Because of these camouflaging features, even experienced gastroenterologists can miss them due to visual fatigue or blind spots.

To overcome the challenge of spotting these flat lesions, doctors use real-time Computer-Aided Detection (CADe). As the gastroenterologist withdraws the colonoscope, the CADe software processes the high-definition video feed at speeds of up to 60 frames per second. The AI relies on deep learning algorithms trained on millions of previous colonoscopy images containing verified flat polyps. When the algorithm recognizes the subtle texture changes, mucous caps, or faint vascular disruption characteristic of a flat SSP, it immediately alerts the physician. This alert usually appears as a brightly colored bounding box on the monitor directly over the suspicious tissue, sometimes accompanied by a subtle audio cue.

Once the CADe system successfully flags a flat lesion, the gastroenterologist utilizes Computer-Aided Diagnosis (CADx) to determine what kind of polyp it is. In the past, a physician had to rely solely on their training to visually guess the polyp's pathology or wait days for a lab biopsy result. CADx streamlines this by acting as an optical biopsy tool in real time.

To use CADx, the doctor often switches the colonoscope to a specialized lighting mode, such as Narrow Band Imaging (NBI), and moves the camera closer to the flat polyp. The CADx algorithm then analyzes the microscopic surface patterns and blood vessel structures of the tissue. Within seconds, the AI outputs a probability score on the screen, differentiating between harmless hyperplastic polyps and dangerous sessile serrated polyps or adenomas.

The integration of CADe and CADx into the endoscopy suite directly impacts patient care during the procedure. When the AI confirms that a flat lesion is likely a sessile serrated polyp, the gastroenterologist knows immediately that the tissue must be completely removed, often using specialized techniques like endoscopic mucosal resection (EMR) to ensure no flat margins are left behind. Conversely, if the AI diagnoses a tiny lesion in the lower colon as a benign hyperplastic polyp, the doctor may safely leave it alone, saving time and reducing the cost of unnecessary pathology lab tests.

Ultimately, real-time AI standardizes the quality of colonoscopies. By eliminating the variables of human eye fatigue and distraction, CADe and CADx empower gastroenterologists to consistently find and accurately diagnose flat sessile serrated polyps, stopping the progression of colorectal cancer before it can begin.