A novel perceptual two layer image fusion using deep learning for imbalanced COVID-19 dataset

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• COVID-19, a global health threat, frequently causes lung infections.
• Chest X-ray (CXR) is a widely accessible diagnostic tool for lung diseases.
• Deep learning (DL) offers potential for analyzing large volumes of medical images.

Purpose of the Study:

• To propose a novel perceptual two-layer image fusion technique using DL for COVID-19 CXR analysis.
• To enhance the informativeness of CXR images for better COVID-19 diagnosis.
• To address challenges with imbalanced COVID-19 datasets.

Main Methods:

• A hybrid decomposition and fusion approach combining Nonsubsampled Contourlet Transform (NSCT) and CNN_VGG19 was developed.
• The method utilizes DL as a feature extractor for CXR images.
• A dataset of 87 COVID-19 confirmed CXR images from 25 cases was used for evaluation.

Main Results:

• The proposed algorithm effectively generates reliable CXR images from imbalanced COVID-19 datasets.
• Fused images exhibit richer features and characteristics compared to the original dataset.
• The NSCT + CNN_VGG19 algorithm demonstrated superior performance in key medical image fusion metrics (Q^MI, PSNR, SSIM).

Conclusions:

• A novel DL-based image fusion algorithm (NSCT + CNN_VGG19) was successfully developed for imbalanced COVID-19 CXR datasets.
• The proposed method significantly enhances image features for improved diagnostic utility.
• Experimental results confirm the superiority of the NSCT + CNN_VGG19 algorithm over existing fusion techniques.