Automated Detection of Anatomical Landmarks During Colonoscopy Using a Deep Learning Model

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Gastroenterology

Background:

• The ileocecal valve (ICV) and appendiceal orifice (AO) are crucial landmarks for confirming complete colonoscopy.
• Automated identification of these landmarks can enhance documentation and quality assessment.

Purpose of the Study:

• To develop and validate a deep convolutional neural network (DCNN) model for automated identification of ICV and AO.
• To assess the model's ability to differentiate these landmarks from normal mucosa and colorectal polyps.

Main Methods:

• A DCNN classification model was trained and validated using 25,444 frames from 318 colonoscopy videos.
• The dataset included images of AO, ICV, normal mucosa, and polyps.
• The model was tested on separate data sets to evaluate its performance.

Main Results:

• The DCNN model identified AO and ICV in 85.7% of patients.
• The model achieved 86.4% accuracy in differentiating AO/ICV from normal mucosa.
• It distinguished polyps from normal mucosa with 88.6% accuracy.

Conclusions:

• The developed DCNN model serves as a novel tool for automated identification of AO and ICV during colonoscopy.
• The model reliably differentiates anatomical landmarks from surrounding tissues and polyps.
• Implementation in automated reporting and quality auditing can enhance colonoscopy reporting quality.