Classification of pediatric video capsule endoscopy images for small bowel abnormalities using deep learning models.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in pediatrics
• Gastroenterology

Background:

• Video capsule endoscopy (VCE) is a noninvasive diagnostic tool for small bowel abnormalities in children.
• Manual review of VCE images is time-consuming and labor-intensive.
• Development of automated methods is crucial for efficient VCE image analysis.

Purpose of the Study:

• To develop and evaluate deep learning models for automatic classification of small bowel lesions.
• To assess the performance of different deep learning architectures using pediatric VCE images.

Main Methods:

• Retrospective analysis of 2298 VCE images from 162 pediatric patients.
• Training and evaluation of four deep learning models: DenseNet121, VGG-16, ResNet50, and Vision Transformer.
• Performance assessment using accuracy, precision, recall, F1-score, and AU-ROC, with lesion visualization via Grad-CAM.

Main Results:

• DenseNet121 and ResNet50 achieved high classification performance, with accuracies of 90.6% and 90.5%, respectively.
• AU-ROC values for DenseNet121 and ResNet50 were 93.7% and 93.4%, respectively.
• Abnormal lesions were detected in 93 children, with abdominal pain being the most common indication for VCE.

Conclusions:

• Deep learning models effectively classify lesions in pediatric VCE images.
• The developed AI tool contributes to more accurate diagnoses.
• This approach enhances diagnostic efficiency in pediatric gastroenterology.