GastroFuse-Net: an ensemble deep learning framework designed for gastrointestinal abnormality detection in endoscopic...

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Gastroenterology

Background:

• Manual interpretation of endoscopic images for gastrointestinal disease diagnosis is labor-intensive, time-consuming, and prone to variability.
• Accurate and automated diagnosis of gastrointestinal conditions from endoscopic images remains a significant challenge, even for experienced specialists.

Purpose of the Study:

• To develop and evaluate a specialized Convolutional Neural Network (CNN) architecture, GastroFuse-Net, for automated recognition of human gastrointestinal diseases from endoscopic images.
• To improve the efficiency and consistency of gastrointestinal disease diagnosis by leveraging deep learning techniques.

Main Methods:

• A novel CNN-based architecture, GastroFuse-Net, was designed by integrating features from shallow and deep CNN models to capture diverse image representations.
• The Kvasir dataset, comprising endoscopic images classified by anatomical structures, diseases, and surgical operations, was utilized for model training and validation.
• Performance was assessed using metrics including precision, recall, specificity, F1-score, Mathew's Correlation Coefficient (MCC), and accuracy.

Main Results:

• GastroFuse-Net demonstrated exceptional performance on the Kvasir dataset.
• The model achieved a precision of 0.985, recall of 0.985, specificity of 0.984, F1-score of 0.997, MCC of 0.982, and an overall accuracy of 98.5%.

Conclusions:

• GastroFuse-Net shows significant promise as an automated tool for diagnosing gastrointestinal diseases from endoscopic images.
• The proposed deep learning approach offers a highly accurate and efficient alternative to manual diagnostic procedures, potentially reducing variability and improving patient care.