GI Genius and AI in the clinic: what do the algorithms find?

What GI Genius actually does during an investigation

GI Genius is used as AI-assisted support during endoscopic procedures, where the system helps point out findings that might otherwise be overlooked. It is therefore a good example of how AI can improve attention and detection in a limited clinical situation. PMID 31585124 PMID 24693890

Precisely that demarcation is important. The value does not come from AI becoming a new doctor. The value comes from the fact that it can elevate a certain part of an investigation if used correctly. PMID 31585124 PMID 24693890

What it can mean for the patient

For the patient, the most relevant question is whether the technology can contribute to better quality and less risk of overlooked findings. It is a more down-to-earth benefit than blanket promises of digital transformation. PMID 24693890 PMID 31775168

At the same time, it is important to know that the gain still depends on the whole setting: equipment, workflow, the clinical professional and how the output is interpreted. PMID 24693890 PMID 31775168

What AI still doesn't solve

AI cannot take responsibility for the entire patient process, interpret all findings in isolation or replace clinical assessment, history and follow-up. This means that better software does not remove the need for human expertise. PMID 31775168 Sundhedsstyrelsen

This is also why concrete cases are so useful. They show both the potential and the limits, instead of letting AI appear as a magical layer on top of healthcare. PMID 31775168 Sundhedsstyrelsen

Why concrete AI tools are more interesting than general hype

Readers benefit more from a concrete example like GI Genius than from yet another broad article about AI being the future. Concrete cases make it possible to assess utility, risk and expectation in a much more sober way. Sundhedsstyrelsen PMID 36306801

These kinds of explanations build trust: less hype, more delineation and better understanding of where the technology actually works. Sundhedsstyrelsen PMID 36306801

Adenoma Detection Rate (ADR) and machine vision during colonoscopy

In colorectal cancer screening and preventive gastroenterology, one benchmark stands supreme: Adenoma Detection Rate (ADR) — the percentage of patients in whom the endoscopist identifies and resects at least one precancerous polyp. PMID 31585124 PMID 24693890

Longitudinal research published in The New England Journal of Medicine demonstrates that for every 1% increase in an endoscopist's ADR, the patient's subsequent risk of developing interval colorectal cancer drops by 3%, while the risk of fatal bowel cancer drops by 5%. PMID 31585124 PMID 24693890

Medtronic GI Genius is a computer-aided detection (CADe) platform powered by deep convolutional neural networks. It scans endoscopic video streams at 25-30 frames per second, highlighting suspected lesions with green bounding boxes in real time. In multicenter randomized controlled trials (Repici et al.), GI Genius increased ADR from 40.4% to 54.8% — an absolute increase of over 14 percentage points. PMID 31585124 PMID 24693890

What the algorithm detects — and where optical blind spots remain

The greatest diagnostic value of AI machine vision lies in identifying subtle, flat sessile serrated lesions (SSL) in the proximal right colon. These lesions blend into the mucosa, often masked by a thin mucus cap, yet account for up to 30% of interval colorectal cancers. PMID 31585124 PMID 31775168 PMID 36306801

Nevertheless, the algorithm is bound by optical physics: it can only analyze what the camera physically illuminates. It cannot peer behind haustral folds, and it cannot see through liquid debris or residual stool. PMID 31585124 PMID 31775168 PMID 36306801

A Boston Bowel Preparation Scale (BBPS) score under 6 renders even state-of-the-art neural networks ineffective. Diligent patient bowel preparation remains the non-negotiable foundation of diagnostic accuracy. PMID 31585124 PMID 31775168 PMID 36306801

Colorectal screening protocols and AI deployment in clinical practice

In national European screening paradigms, citizens aged 50-74 undergo biennial fecal immunochemical testing (FIT). A positive occult blood result triggers expedited referral to high-volume endoscopy suites. PMID 24693890 Sundhedsstyrelsen

The rollout of AI detection systems like GI Genius across major hospital departments and private specialty clinics (such as Aleris and Capio) serves as a persistent, fatigue-free 'second observer', maintaining high detection sensitivity throughout long procedural shifts. PMID 24693890 Sundhedsstyrelsen

The table below contrasts standard white-light colonoscopy against AI-augmented endoscopy. PMID 24693890 Sundhedsstyrelsen