Three Italian endoscopy centers using GI Genius
Verified evidenceCreator to judgeThat the endoscopist must be the only visual observer capable of flagging a lesion during withdrawal.
Clinical operations · Italy
Collections: Human still decides · Regulated autonomy
Executive brief
The model changed the attention workflow rather than the clinical decision right; value came from watching every frame without slowing the procedure.
AI value · Adenomas detected per colonoscopy
Verified1.07 in the CADe group; incidence-rate ratio 1.46, with no increase in withdrawal time
Randomized detection endpoint; the trial did not establish reduced colorectal-cancer mortality, and much of the uplift was in small lesions.
Before
Inspect the live colonoscopy feed and visually identify suspected lesions. → Determine whether removed tissue is neoplastic.
After
Processes the live video and superimposes a green box over a suspected lesion. → Inspects the highlighted area and decides whether to remove tissue.
Human boundary
The AI draws attention; the endoscopist decides whether a finding is real and whether to resect.
Why it matters
That the endoscopist must be the only visual observer capable of flagging a lesion during withdrawal.
Before
How the work ran before the change.
Endoscopist
Inspect the live colonoscopy feed and visually identify suspected lesions.
ControlClinical judgment and minimum withdrawal time
Pathologist
Determine whether removed tissue is neoplastic.
ControlHistopathology reference standard
What changed
That the endoscopist must be the only visual observer capable of flagging a lesion during withdrawal.
Decision rightHuman moves from creator to judge
After
How the same work runs now.
GI Genius CADe
Processes the live video and superimposes a green box over a suspected lesion.
ControlReal-time computer-aided detection
Endoscopist
Inspects the highlighted area and decides whether to remove tissue.
ControlPhysician retains procedural decision rights
Exception path
The physician ignores false alerts, inspects missed or ambiguous tissue directly, and relies on histopathology for final classification.
Work removed
Decision authority
The AI draws attention; the endoscopist decides whether a finding is real and whether to resect.
Before
Endoscopist
Inspect the live colonoscopy feed and visually identify suspected lesions.
Control: Clinical judgment and minimum withdrawal time
Pathologist
Determine whether removed tissue is neoplastic.
Control: Histopathology reference standard
After
GI Genius CADe
Processes the live video and superimposes a green box over a suspected lesion.
Control: Real-time computer-aided detection
Endoscopist
Inspects the highlighted area and decides whether to remove tissue.
Control: Physician retains procedural decision rights
Work that left the path
Human role before
The endoscopist acted as the sole real-time visual detector.
Human role after
The endoscopist remains the procedural decision maker but works with an always-on secondary observer.
AI roleDeep-learning computer vision that flags suspected colorectal lesions on the live display.
0.71 mean adenomas per colonoscopy in the standard-colonoscopy control group→1.07 in the CADe group; incidence-rate ratio 1.46, with no increase in withdrawal time
September-November 2019 randomized trial · 685 subjects at three Italian centers
Randomized detection endpoint; the trial did not establish reduced colorectal-cancer mortality, and much of the uplift was in small lesions.
Anti-pattern
Calling every AI box a diagnosis or extrapolating detection uplift to mortality.
Questions
Portability conditions
Reputation risk
low
Evidence and authority
Current · updated
1 peer reviewed; publication outcomes are verified.
Bundle 1.0.0 · reviewed 2026-08-23 · stable ID fb7af1efc542b6d1
University of California, San Francisco and UC Berkeley · Creator to judge
Attempted-speech brain signals can produce real-time text, audio, and facial animation as a faster channel.
Johns Hopkins Health System · Creator to judge
An alert only changes care when a clinician confirms it quickly.
Google · Creator to judge
Large migrations can combine machine-found locations, LLM edits, automated checks, and engineer review.
Rising Academies schools · Creator to judge
Students can receive adaptive, low-bandwidth math practice without a scarce one-to-one tutor.