AI-KODA score application for cleanliness assessment in video capsule endoscopy frames

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Video capsule endoscopy (VCE) lacks automated cleanliness assessment methods.
• The KOrea-CanaDA (KODA) scoring system is used for VCE frame evaluation.
• Developing automated tools for VCE analysis is crucial for clinical efficiency.

Purpose of the Study:

• To automate VCE frame scoring using the KODA system via a mobile application (AI-KODA).
• To assess the inter-rater and intra-rater reliability of the KODA score using the AI-KODA application.
• To evaluate the efficacy of the AI-KODA application for VCE cleanliness assessment.

Main Methods:

• Selected 1539 sequential and 634 random frames from 28 VCE patient videos.
• Three gastroenterology fellows rated 2173 frames twice using the AI-KODA app.
• Inter-class correlation coefficients (ICCs) were used to assess reliability.
• Random forest and k-nearest neighbors classifiers were employed for automatic score classification.

Main Results:

• High inter-rater reliability (excellent to good) for sequential frames and excellent for random frames.
• Good to moderate intra-rater reliability for sequential frames and excellent for random frames.
• Classification accuracy reached up to 61% (random forest) and 62.38% (k-nearest neighbors).

Conclusions:

• AI-KODA successfully automates VCE frame scoring based on the KODA system.
• The application offers a user-friendly and time-saving solution for VCE cleanliness assessment.
• AI-KODA demonstrates potential for both research and clinical applications in VCE analysis.