Automated Detection of Gastrointestinal Diseases Using Resnet50*-Based Explainable Deep Feature Engineering Model...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Gastrointestinal disease diagnosis relies heavily on endoscopic imaging.
• Accurate and early detection of gastrointestinal diseases is crucial for effective treatment.
• Current diagnostic methods can be limited by human interpretation and variability.

Purpose of the Study:

• To develop a novel convolutional neural network (CNN), ResNet50*, for detecting gastrointestinal diseases.
• To create an explainable deep feature engineering (DFE) model integrated with ResNet50* for enhanced diagnostic capabilities.
• To evaluate the performance and generalizability of the proposed ResNet50* model across multiple endoscopic image datasets.

Main Methods:

• Development of ResNet50*, a CNN variant with residual blocks and a pooling-based attention mechanism.
• Implementation of a four-stage DFE model: feature extraction, iterative feature selection, shallow classification (kNN, SVM), and information fusion.
• Utilized Gradient-weighted Class Activation Mapping (Grad-CAM) for feature visualization and selection.
• Trained and validated the model on augmented Kvasir, Kvasir version 2, and wireless capsule endoscopy (WCE) datasets.

Main Results:

• The ResNet50* model achieved over 92% classification accuracy on all tested datasets.
• A remarkable accuracy of 99.13% was recorded for the wireless capsule endoscopy (WCE) dataset.
• The DFE model produced 14 distinct outcomes, with the final decision selected based on greedy algorithm accuracy.
• Consistent performance across diverse endoscopic image datasets confirmed model generalizability.

Conclusions:

• The proposed ResNet50* model demonstrates superior classification performance for gastrointestinal diseases.
• The explainable DFE model enhances diagnostic accuracy and provides interpretable results.
• The developed architecture shows significant potential for improving automated detection of gastrointestinal conditions in clinical practice.