GestroNet: A Framework of Saliency Estimation and Optimal Deep Learning Features Based Gastrointestinal Diseases...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastrointestinal Diseases

Background:

• Manual diagnosis of gastrointestinal diseases is time-consuming, costly, and requires expert interpretation.
• Computerized methods are needed to assist clinicians by providing a second opinion.
• Accurate segmentation of infected regions is crucial for reliable feature extraction and classification in automated systems.

Purpose of the Study:

• To propose an automated framework for segmenting and classifying gastrointestinal tract (GIT) diseases.
• To address the challenge of accurate infected region segmentation in automated diagnostic systems.
• To enhance the accuracy of GIT disease classification using deep learning and optimized feature selection.

Main Methods:

• Image preprocessing with a contrast enhancement technique.
• Deep saliency maps for segmenting infected regions.
• Transfer learning with a fine-tuned MobileNet-V2 model, hyperparameters optimized by Bayesian optimization (BO).
• Feature extraction using average pooling, followed by feature selection using a hybrid whale optimization algorithm.
• Classification using an extreme learning machine (ELM) classifier.

Main Results:

• The proposed framework achieved high classification accuracies: 98.20% on Kvasir 1, 98.02% on Kvasir 2, and 99.61% on CUI Wah datasets.
• Demonstrated significant improvement in accuracy compared to existing methods.
• Successfully automated the segmentation and classification of gastrointestinal infections.

Conclusions:

• The developed automated framework effectively segments and classifies gastrointestinal diseases with high accuracy.
• The combination of deep saliency maps, Bayesian optimization, and whale optimization algorithm for feature selection provides a robust approach.
• This AI-driven system shows potential for assisting medical professionals in diagnosing gastrointestinal infections more efficiently.