Deep learning-based prediction model for diagnosing gastrointestinal diseases using endoscopy images.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Gastrointestinal (GI) infections are prevalent globally.
• Current diagnostic methods like colonoscopy and wireless capsule endoscopy (WCE) are time-consuming and prone to human error.
• Automated AI-based diagnostic tools are crucial for improving early detection and patient care.

Purpose of the Study:

• To develop and evaluate an AI-based system for the early diagnosis of GI diseases.
• To enhance diagnostic accuracy using a convolutional neural network (CNN).
• To assess the performance of different CNN models, including transfer learning approaches.

Main Methods:

• Trained various CNN models (baseline, VGG16, InceptionV3, ResNet50) on the KVASIR dataset.
• Utilized n-fold cross-validation and data augmentation strategies.
• Evaluated model performance on a test set of 1200 images, including healthy and diseased states (polyps, ulcerative colitis, esophagitis).

Main Results:

• The ResNet50 pre-trained CNN model achieved the highest accuracy (approx. 99.80% on training, 99.50% on validation, and 99.16% on test sets).
• The ResNet50 model demonstrated high precision (100%) and recall (approx. 99%) in diagnosing GI diseases.
• The proposed ResNet50 model outperformed existing systems in diagnostic accuracy.

Conclusions:

• AI-based prediction models, particularly CNNs like ResNet50, significantly improve diagnostic accuracy for GI diseases.
• The developed model aids in the early detection of gastrointestinal polyps, ulcerative colitis, and esophagitis.
• The prediction model is publicly available for further research and application.