Evaluation of Computer-aided Detection for Identifying Missed Gastric Cancer After Endoscopic Submucosal Dissection.

Key Takeaways for GI Nurses

• Computer-aided detection (CADe) for missed gastric cancers (MGCs) after endoscopic submucosal dissection (ESD) showed a per-lesion sensitivity of only 15.0%, statistically similar to the 13.8% sensitivity of experienced endoscopists—meaning the technology is not yet reliably catching these difficult-to-detect post-ESD lesions.
• CADe demonstrated a significantly faster per-image diagnostic time (0.03 seconds vs. 2.86 seconds for endoscopists), highlighting its potential as a rapid screening aid rather than a standalone diagnostic tool.
• Endoscopists outperformed CADe in specificity (94.8% vs. 88.5%) and positive predictive value (8.3% vs. 4.4%), meaning CADe currently generates more false-positive alerts, which nursing staff should anticipate during procedures.
• This study underscores that surveillance endoscopy after ESD remains a high-vigilance task requiring careful visual inspection by the entire care team, as MGCs are inherently difficult to identify regardless of whether human or artificial intelligence is doing the looking.

Clinical Relevance

For nurses working in endoscopy suites that perform surveillance esophagogastroduodenoscopy (EGD) after ESD, this study offers a sobering but important reality check on the current limitations of AI-assisted detection technology. While CADe systems are often marketed as a solution to reduce missed lesions, this research demonstrates that in the specific and clinically challenging context of post-ESD scar surveillance, CADe performance is roughly equivalent to human detection—and importantly, it is not yet a substitute for thorough endoscopist assessment. Nurses assisting during these procedures should understand that CADe alerts are not infallible and that vigilant support of the endoscopist, including careful attention to positioning, mucosal cleansing, and adequate visualization of scar tissue, remains essential to lesion detection regardless of technological assistance.

The finding that CADe produces more false positives (lower specificity and positive predictive value) than endoscopists has direct workflow implications. Nurses should be prepared for the possibility that AI-flagged areas may prompt additional biopsies or closer inspection that ultimately reveal benign findings, which can extend procedure time and require additional specimen handling and patient counseling. Understanding this trade-off allows nursing staff to better anticipate procedural flow when CADe systems are integrated into practice, and to communicate realistic expectations to patients regarding what AI assistance can and cannot guarantee in cancer surveillance.

From a professional development standpoint, this study reinforces the importance of nurses staying informed about the evolving landscape of AI-assisted endoscopy tools. As these technologies continue to be studied and refined, nurses play a critical role in observing real-world performance, documenting outcomes, and providing feedback that can inform quality improvement initiatives within their endoscopy units. Understanding the nuanced strengths and limitations of CADe—such as its speed advantage but current sensitivity limitations in detecting subtle post-ESD recurrences—equips nurses to be more effective advocates for patient safety and informed participants in decisions about technology adoption.

Bottom Line

Nurses should recognize that current CADe technology, while fast, does not yet outperform experienced endoscopists in detecting missed gastric cancers after ESD, meaning careful clinical vigilance, thorough scar inspection, and strong endoscopist-nurse teamwork remain the most reliable safeguards during post-ESD surveillance endoscopy.