Software Analysis of Colonoscopy Videos Enhances Teaching and Quality Metrics.

Area of Science:

• Endoscopic imaging
• Machine learning applications in medicine
• Gastrointestinal diagnostics

Background:

• Colonoscopy quality is crucial for accurate diagnosis and patient outcomes.
• Objective assessment of colonoscopy luminal visualization is challenging.
• Existing quality indicators in endoscopy require enhancement.

Purpose of the Study:

• To evaluate the efficacy of machine learning algorithms in predicting colonoscopy luminal image quality.
• To develop a novel, automated metric for assessing endoscopic visualization.
• To explore the potential of AI in improving endoscopy training and quality assurance.

Main Methods:

• Colonoscopy videos from a randomized controlled trial were analyzed.
• An algorithm utilizing steerable filters-based metrics (SFIL) assessed luminal visualization.
• SFIL scores were compared against two independent clinician assessments (C1, C2) using Goodman and Kruskal's gamma (G).

Main Results:

• A total of 412 colonoscopy video clips were analyzed.
• SFIL scores showed agreement with C1 in 45% and C2 in 42% of cases.
• Significant correlations were found between SFIL and clinician assessments (G = 0.644 for C1, G = 0.734 for C2, both p < 0.005).

Conclusions:

• Machine learning algorithms, specifically SFIL, can accurately recognize the quality of luminal visualization during colonoscopy.
• This AI-driven approach shows promise for enhancing colonoscopy training programs.
• The developed algorithm can serve as a valuable metric for objective quality assessment in endoscopy.