Artificial Intelligence and Capsule Endoscopy: Automatic Detection of Small Bowel Blood Content Using a Convolutional...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Capsule endoscopy is vital for diagnosing obscure gastrointestinal bleeding.
• Manual review of capsule endoscopy images is time-consuming and can miss lesions.
• Improving the diagnostic yield of capsule endoscopy is crucial.

Purpose of the Study:

• To develop a deep learning algorithm for automatic detection of blood in capsule endoscopy.
• To enhance the diagnostic accuracy and efficiency of capsule endoscopy exams.

Main Methods:

• A convolutional neural network was trained on 22,095 capsule endoscopy images.
• The network's performance was validated against specialist classifications.
• Key performance metrics including sensitivity, specificity, accuracy, and precision were calculated.

Main Results:

• The AI model achieved high accuracy (98.5%) and precision (98.7%) in detecting blood.
• Sensitivity and specificity were also excellent at 98.6% and 98.9%, respectively.
• The AI processed the validation dataset rapidly (24 seconds).

Conclusions:

• An artificial intelligence tool effectively detects luminal blood in capsule endoscopy.
• This AI has the potential to significantly improve diagnostic accuracy for obscure small bowel bleeding.