Deep ensemble learning for gastrointestinal diagnosis using endoscopic image classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Gastrointestinal tract (GIT) abnormalities are a leading cause of mortality.
• Diagnosing GIT disorders via endoscopy presents challenges like limited annotated images and poor image quality.
• Computer-aided diagnosis (CAD) systems are crucial for assisting gastroenterologists.

Purpose of the Study:

• To develop a transfer learning-based deep ensemble model for accurate classification of endoscopic GIT images.
• To enhance the accuracy and efficiency of diagnosing gastrointestinal disorders.
• To address limitations in existing CAD systems for endoscopy.

Main Methods:

• A deep ensemble model was formulated using a weighted voting ensemble of NasNet-Mobile and EfficientNet.
• Regions of interest (affected areas) were extracted from endoscopic images.
• Cross-dataset evaluation was performed using HyperKvasir, Kvasir v1, and Kvasir v2 datasets.

Main Results:

• The proposed model achieved high accuracy, with 97.83% on Kvasir v1 and 98.45% on Kvasir v2.
• Performance metrics included accuracy, precision, recall, Area Under Curve (AUC), and F1 score.
• The model demonstrated superior performance compared to existing transfer learning models.

Conclusions:

• Transfer learning is a powerful approach for developing robust CAD systems in gastroenterology.
• The proposed deep ensemble model effectively classifies GIT disorders from endoscopic images.
• This framework offers a promising tool for improving diagnostic accuracy and patient outcomes.