Deep Q-Learning for Gastrointestinal Disease Detection and Classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Stomach ulcers are a prevalent gastrointestinal disease requiring early diagnosis.
• Timely detection and treatment are crucial to mitigate serious health risks associated with stomach ulcers.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for accurate classification and segmentation of stomach ulcers.
• To enhance diagnostic capabilities for gastrointestinal diseases through advanced AI models.

Main Methods:

• A classification model using Convolutional Neural Networks (CNN) integrated with Q-learning for stable and accurate ulcer identification.
• A segmentation model employing an attention-based U-Net architecture to precisely delineate ulcer lesion regions.
• Performance evaluation across multiple datasets including Kvasir, Nerthus, CVC-ClinicDB, and a private POF dataset.

Main Results:

• The classification framework achieved 99.08% accuracy on the Kvasir dataset and 100% on the Nerthus dataset.
• The segmentation framework demonstrated high performance with 98.09% accuracy on Kvasir, 99.77% on Nerthus, 98.49% on CVC-ClinicDB, and 99.13% on the private dataset.
• The proposed models outperformed existing methods in stomach ulcer classification and segmentation.

Conclusions:

• The integrated deep learning approach offers a highly effective solution for stomach ulcer diagnosis.
• This AI-driven method shows significant potential for improving the early detection and management of gastrointestinal diseases.