A Robust Hybrid Preprocessing Framework Integrating DAE, BM3D, Median Filtering, and CLAHE for Enhanced Pneumonia Chest X-Ray Analysis

Abstract

For the identification of pneumonia using CXR imaging, image quality is potentially impacted by noise, low contrast, and the presence of acquisition artifacts that make it difficult to visualize pertinent or clinically relevant features. To improve the quality of CXR images associated with pneumonia diagnoses, this study recommends the use of a robust hybrid preprocessing framework that utilizes learning-based and classical preprocessing techniques for noise reduction, structural preservation, and increasing contrast, and combines these techniques into one model. The learning-based method used is a Denoising Autoencoder (DAE). The classical methods used include Median Filtering, Block-Matching and 3D Filtering (BM3D), and Contrast-Limited Adaptive Histogram Equalization (CLAHE). To evaluate how effectively each preprocessing technique or combination of techniques can enhance pneumonia CXR images, CXR images from publicly available pneumonia CXR datasets were used in this study for evaluation. The performance evaluation used a range of quantitative metrics that included Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM), Mean Squared Error (MSE), Entropy, and Contrast Improvement Index (CII). The results indicate that the proposed hybrid approach is superior to all other preprocessing methods when it comes to creating high-quality images while preserving structural integrity and improving contrast. Therefore, it is concluded that the proposed hybrid preprocessing framework provides a reliable base for improved visual analysis of pneumonia CXR images and enhanced performance from deep learning-based pneumonia detection applications using CXR images.