A two-phase transfer learning framework for gastrointestinal diseases classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Gastrointestinal (GI) disorders are prevalent and can be debilitating.
• Artificial intelligence (AI), specifically computer vision, shows promise in medical image analysis.
• Deep convolutional neural networks (CNNs) excel at identifying complex patterns in images.

Purpose of the Study:

• To develop and evaluate an AI-driven, two-phase transfer learning framework for enhanced GI disease identification from endoscopic images.
• To assist gastroenterologists in achieving more efficient and accurate diagnoses.
• To fine-tune pre-trained models and train custom CNNs for comparative performance analysis.

Main Methods:

• Utilized a two-phase transfer learning framework.
• Fine-tuned three pre-trained models (Xception, InceptionResNetV2, VGG16) on annotated GI endoscopic image datasets.
• Constructed and trained two custom CNNs for comparison.
• Evaluated performance across four distinct classification tasks.

Main Results:

• The InceptionResNetV2 architecture achieved high accuracy across tasks: 85.7% (8-category GI classification), 97.6% (3-disease classification), 99.5% (polyp identification), and 74.2% (esophagitis severity).
• The transfer learning framework demonstrated consistent and generalized performance.
• AI effectively identified GI diseases from unseen endoscopic images.

Conclusions:

• The proposed two-phase transfer learning framework is effective for clinical application in GI disease identification.
• AI-assisted diagnosis can enhance early detection and treatment of GI disorders.
• The study highlights the potential of AI in improving gastroenterological diagnostics.