A deep learning based framework for the classification of multi- class capsule gastroscope image in gastroenterologic...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Gastroscopy is a standard diagnostic tool for gastric conditions.
• Capsule endoscopy offers a novel screening method for gastric diseases.
• Challenges in capsule endoscopy include image quality and operator variability, impacting diagnostic accuracy.

Purpose of the Study:

• To develop an automated method for classifying capsule gastroscope images into three categories: normal, chronic erosive gastritis, and ulcer images.
• To prevent misdiagnosis of high-risk factors for carcinogenesis, such as atrophic gastritis.
• To create a deep learning framework utilizing transfer learning for enhanced image classification.

Main Methods:

• A deep learning framework based on transfer learning was proposed.
• Pre-trained models (VGG-16, ResNet-50, Inception V3) were fine-tuned for the classification task.
• A dataset of 380 images per category was used, split into 70% training and 30% testing sets.

Main Results:

• The VGG-16 model achieved the highest accuracy of 94.80% in classifying capsule gastroscopic images.
• The proposed approach demonstrated respectable specificity and accuracy in classifying images.
• The system effectively distinguished between normal, chronic erosive gastritis, and ulcer images.

Conclusions:

• The developed deep learning framework can improve the accuracy of capsule gastroscope image classification.
• This AI-assisted approach can mitigate diagnostic errors stemming from image quality and human factors.
• The study highlights the potential of deep learning to enhance gastritis diagnosis and improve overall gastroscopy standards.