Deep learning-based localization and lesion detection in capsule endoscopy for patients with suspected small-bowel...

Area of Science:

• Gastroenterology
• Medical Imaging
• Artificial Intelligence

Background:

• Small-bowel capsule endoscopy (SBCE) is crucial for diagnosing obscure gastrointestinal bleeding but is labor-intensive and subjective.
• Current artificial intelligence (AI) solutions often lack simultaneous small-bowel (SB) localization and abnormality detection capabilities.

Purpose of the Study:

• To develop and validate an AI model for automated SB localization within the gastrointestinal tract.
• To create an AI model capable of accurately detecting various SB abnormalities from capsule endoscopy images.

Main Methods:

• An AI model was trained on a large dataset of 87,005 capsule endoscopy images for organ localization (stomach, SB, colon).
• A separate dataset of 28,405 SBCE images was used to train the AI for detecting SB abnormalities like erosions, ulcers, angiodysplasia, and bleeding.
• Diagnostic performance was evaluated by comparing AI-assisted reading with conventional reading of 32 SBCE videos.

Main Results:

• The AI model demonstrated high accuracy (>97%) and AUC (>0.99) for SB localization.
• Excellent performance was achieved in detecting SB abnormalities, with accuracies ranging from 99.4% to 99.9% and AUCs up to 0.99.
• AI-assisted reading significantly reduced interpretation time (8.7 minutes vs. 53.9 minutes) with comparable diagnostic accuracy to conventional methods.

Conclusions:

• The developed AI model effectively decreases SBCE reading time.
• AI integration into SBCE reading provides efficient and reliable detection of small-bowel abnormalities.
• This AI tool shows promise for enhancing the diagnostic workflow in capsule endoscopy.