Artificial intelligence and capsule endoscopy: automatic detection of vascular lesions using a convolutional neural...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Capsule endoscopy (CE) is the primary tool for evaluating obscure gastrointestinal bleeding.
• Small intestinal vascular lesions are common but challenging to diagnose accurately from CE images.
• Current CE reading is time-consuming and error-prone, necessitating advanced diagnostic tools.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) model for identifying and differentiating small intestinal vascular lesions.
• To assess the model's ability to distinguish lesions based on their hemorrhagic potential.
• To improve the accuracy and efficiency of capsule endoscopy interpretation.

Main Methods:

• A CNN model was developed using a database of 11,588 capsule endoscopy images.
• Images included normal mucosa, red spots, and angiectasia/varices, classified by hemorrhagic risk.
• The dataset was divided into training and testing sets for model evaluation.

Main Results:

• The CNN model achieved high accuracy in detecting vascular lesions, with 91.8% sensitivity and 95.9% specificity.
• Specific detection rates for red spots were 97.1% sensitivity and 95.3% specificity.
• Angiectasia/varices detection showed 94.1% sensitivity and 95.1% specificity, with a rapid processing speed of 145 frames/sec.

Conclusions:

• This is the first CNN-based model capable of accurately detecting and differentiating enteric vascular lesions by hemorrhagic risk.
• CNN-assisted capsule endoscopy reading offers a promising approach to enhance diagnostic accuracy and efficiency.
• The developed algorithm has the potential to significantly improve patient diagnosis and management of gastrointestinal bleeding.