GI-ScreenNet v2: A Modular Framework for Gastrointestinal Disease Detection Based on an Integrated Transfer Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Gastrointestinal diseases represent a major global health concern.
• Wireless capsule endoscopy (WCE) is crucial for early screening but generates large datasets.
• Manual analysis of WCE images is labor-intensive and error-prone, hindering clinical adoption of AI solutions due to inflexibility and complexity.

Purpose of the Study:

• To develop a flexible and efficient AI framework for analyzing wireless capsule endoscopy (WCE) images.
• To improve the accuracy and reduce the time required for gastrointestinal disease screening.

Main Methods:

• Introduction of GI-ScreenNet v2, a novel multi-backbone network framework.
• Utilizes ensemble and transfer learning with a standardized interface for arbitrary backbone support.
• Employs a cross-attention mechanism for dynamic feature integration and sophisticated representation learning.

Main Results:

• GI-ScreenNet v2 achieved 94.87% accuracy on the KvasirV2 dataset.
• Demonstrated a 3.31% performance improvement over traditional methods.
• Indicated potential for efficient gastrointestinal screening and AI-assisted diagnostics.

Conclusions:

• Presents a unified framework for gastrointestinal disease detection using AI.
• Features an integrated workflow for dynamic model selection and cross-attention fusion.
• Facilitates the integration of new models and techniques, advancing robust diagnostic systems.