# Poster Session I - A60 A SURVEY EVALUATING USER PERCEPTIONS OF A COMMERCIAL ARTIFICIAL INTELLIGENCE-BASED POLYP DETECTION SYSTEM DURING COLONOSCOPY

## TL;DR

This study evaluates how healthcare professionals perceive a commercial AI tool for detecting polyps during colonoscopies, finding mixed opinions on its effectiveness and impact on workflow.

## Contribution

The study provides novel insights into clinician perceptions and usability challenges of AI-based polyp detection systems in real-world settings.

## Key findings

- CAD EYE was rated as only moderately effective by participants on a 5-point scale.
- Workflow disruption and false positives were the most frequently reported limitations.
- Improved specificity and customizable alerts could enhance user experience and adoption.

## Abstract

Colorectal cancer is a largely preventable disease that can be prevented through detection and removal of precancerous polyps (adenomas), most commonly through colonoscopy. However, despite its effectiveness, finding all polyps remains a challenge, with several adjuncts being proven to assist in polyp detection. Artificial intelligence (AI) systems, such as FUJIFILM’s CAD EYE, have been developed to support real-time polyp detection during colonoscopy treatments. While clinical trials suggest potential benefits, the success of such tools relies primarily on clinician perceptions and usability in routine practice.

This cross-sectional study aimed to assess healthcare professionals’ experiences and perceptions of CAD EYE, with a focus on its acceptance, perceived effectiveness, and its impact on workflow during colonoscopy.

Nineteen healthcare professionals, including gastroenterologists, fellows, and trainees, participated in a structured and confidential survey. Survey responses were analyzed using descriptive statistics and thematic analysis. One participant was excluded due to lack of experience with the software.

Of the 18 surveyed participants, 4 reported regular use of CAD EYE, 5 reported no present use, and 9 reported occasional use. Compared with their past usage, 5 participants had completely stopped using CAD EYE, 5 reported reduced use, and 6 reported increased use. Yet, on a 5-point scale (5 = very effective), CAD EYE was rated an average of 2.4. In terms of workflow, 11 participants indicated that CAD EYE made their work slower, 6 reported it made no difference, and 1 found it faster.

Reasons for use or non-use fell into eight distinct categories: distraction (n = 6), negative workflow impact (n = 3), limited experience (n = 1), ineffective for polyp detection (n = 3), beneficial for polyp detection (n = 3), false positives (n = 5), second-opinion support (n = 4), and noise annoyance (n = 4). CAD EYE was most frequently used during general polyp detection on withdrawal (n = 4), polyp characterization (n = 8), and procedures involving high-risk or special populations (n = 1).

Perceived benefits included improved polyp detection (n = 6), whereas limitations included workflow disruption from visual and audio cues (n = 6), oversensitivity and false positives (n = 7), and a general need for improvement (n = 2).

Although potential clinical benefits in polyp detection, CAD EYE’s acceptance in the one center’s clinical group is limited by perceived inefficiencies and frequent false-positive results. With enhanced specificity and customizable alerts, CAD EYE could significantly improve user experience and its implementation in the field.

None