Colonoscopy assisted by AI helped increase precancerous growth detection and will likely be implemented into future screening procedures, according to an expert.
AI in medicine has been highly anticipated, partially because of its potential to aid physicians. Dennis L. Shung, MD, MHS, PhD, an assistant professor of medicine, director of digital health and director of applied Artificial Intelligence at Yale University, and colleagues conducted a systematic review evaluating the benefits of AI-assisted colonoscopy. They evaluated 44 randomized controlled trials with 36,201 cases and published their results in Annals of Internal Medicine.
Healio spoke with Shung to learn more about the study and AI's potential in cancer prevention.
Healio: Why did you decide to study AI's impact on CRC screening? Why is this important?
Shung: CRC is the third most common cancer in the United States, and polyp detection with AI has been the best studied area of AI application for clinical medicine.
Healio: Will you briefly describe your results and their clinical implications?
Shung: We found that AI for polyp detection increases the detection of precancerous tubular adenomas, decreases miss rates for adenomas and may improve the detection of advanced colorectal neoplasia. This suggests that in a trial setting, AI systems for polyp detection could improve the performance of expert endoscopists.
Healio: What is AI's potential to improve cancer screening or health care more broadly?
Shung: The potential of this narrow system is yet unknown for more meaningful clinical endpoints, such as CRC-related mortality or interval post-colonoscopy colorectal cancer. Higher quality colonoscopies could eventually lead to longer intervals between colonoscopies and more personalized risk assessments for patients for screening regimens. The base technology for endoscopic AI could be integrated together into a platform that could use generative AI to provide an integrated workflow for patients and providers, decreasing unnecessary documentation and providing more time for providers and patients to interact.
Healio: Are there any concerns about using AI in this way in general?
Shung: Implementation of this AI system into endoscopist workflows could lead to deskilling for trainees, distraction for gastroenterologists and, in the short-term, additional costs to providers and patients with an increased number of unnecessary colonoscopies under the current recommendations for colorectal cancer surveillance intervals. In general, the risk of AI deployment without adequate understanding of human-AI interaction and usability concerns may lead to unanticipated consequences and failure modes.
Healio: What is the take-home message for primary care providers?
Shung: For PCPs, the message is that AI systems for polyp detection will likely be integrated into future screening procedures and may change recommended intervals for surveillance colonoscopies.
Healio: Is there anything else you would like to add?
Shung: Any solutions should integrate patient preferences and provider concerns, with the primary focus on the interest of patients. From a technology standpoint, these detection algorithms will eventually be integrated into platforms with large multimodal models that will transform provider workflows.