Deep Learning and Minimally Invasive Endoscopy: Automatic Classification of Pleomorphic Gastric Lesions in Capsule...

Area of Science:

• Gastroenterology
• Artificial Intelligence
• Medical Imaging

Background:

• Capsule endoscopy (CE) is a key tool for gastrointestinal tract evaluation.
• Current diagnostic yield for gastric lesions via CE is suboptimal.
• The application of AI, specifically CNNs, in analyzing CE images for gastric pathology remains underexplored.

Purpose of the Study:

• To develop and validate a CNN-based algorithm for the automatic classification of diverse gastric lesions detected through wireless capsule endoscopy (WCE).
• To assess the diagnostic performance of the developed CNN algorithm in identifying gastric abnormalities.

Main Methods:

• A CNN algorithm was developed using a dataset of 12,918 gastric images from three different CE devices.
• Images included various lesions such as vascular abnormalities, protruding lesions, ulcers, erosions, and normal mucosa.
• The CNN model underwent training and validation, with performance evaluated against expert gastroenterologist consensus.

Main Results:

• The CNN achieved high diagnostic performance: 97.4% sensitivity, 95.9% specificity, 95.0% positive predictive value, and 97.8% negative predictive value for gastric lesions.
• Overall accuracy for lesion detection was 96.6%.
• The algorithm demonstrated rapid image processing at 115 images per second.

Conclusions:

• This study presents the first CNN algorithm capable of automatically detecting pleomorphic gastric lesions in CE.
• The developed AI tool shows significant potential to enhance the diagnostic accuracy and efficiency of gastric evaluations using WCE.