Empirical Mode Decomposition-Based Deep Learning Model Development for Medical Imaging: Feasibility Study for...

Area of Science:

• Artificial Intelligence
• Medical Imaging Analysis
• Signal Processing

Background:

• Accurate medical image classification is crucial for early disease detection.
• Deep learning models show promise but can be further optimized for complex medical datasets.
• Gastrointestinal endoscopic imaging presents unique challenges for automated analysis.

Purpose of the Study:

• To develop and evaluate a novel deep learning framework incorporating two-dimensional Empirical Mode Decomposition (2D EMD) for enhanced multi-class image classification.
• To assess the impact of 2D EMD on the performance of various deep learning architectures in classifying gastrointestinal endoscopic images.
• To explore the potential of this approach for early disease detection in healthcare.

Main Methods:

• A deep learning framework was designed utilizing 2D EMD to decompose medical images into intrinsic mode functions (IMFs).
• The Kvasir dataset, comprising 8000 gastrointestinal endoscopic images across eight classes, was used for validation.
• The efficacy of integrating 2D EMD was evaluated by comparing four established deep learning models (ResNet152, VGG19bn, MobileNetV3L, SwinTransformerV2S) with and without EMD preprocessing.

Main Results:

• Subtracting IMFs derived from 2D EMD consistently improved accuracy, F1-score, and AUC across all evaluated deep learning models.
• Significant performance enhancements were observed, including approximately 9% accuracy increase for ResNet152, 18% for VGG19L, 3% for MobileNetV3L, and 8% for SwinTransformerV2.
• Explainable AI techniques (Grad-CAM) confirmed that the models focused on relevant gastrointestinal regions for accurate predictions.

Conclusions:

• The integration of 2D EMD significantly enhances the performance of deep learning models for gastrointestinal image classification.
• This 2D EMD-based approach offers a promising strategy for improving AI-driven medical image analysis and disease detection.
• The framework demonstrates potential for broader applications in medical imaging and other image classification domains.