Pneumonia screening on chest X-rays with optimized ensemble model

Abstract

Pneumonia is a lung illness that may result from a variety of various viral diseases and may be lethal. It might be difficult to diagnose and treat pneumonia on chest X-ray pictures because pneumonia and other lung illnesses are closely connected. As a result, the present methods for forecasting pneumonia cannot achieve meaningful levels of accuracy. This study aids in identifying diseases from chest X-ray images that don't exhibit the same symptoms as pneumonia. Five stages will be included in the projected model: (a) pre-processing, (b) segmentation, (c) feature extraction, (d) optimal feature selection, and (e) detection. The gathered image is initially pre-processed using Dynamic Histogram Equalization (DHE), followed in this case by median filtering. At the segmentation phase, by utilizing the Enhanced Watershed Segmentation, the ROI is separated from the background region of the pre-processed image. Additionally, in the feature extraction step, features based on Local Gradient Increasing Pattern (LGIP), Harmonic Local Gradient Pattern (hLGP), Improved Local Ternary Pattern (ILTP), and Improved Local Gradient Pattern (ILGP) will be derived from the isolated ROI areas. Using the proposed AHGOA (Archimedes-assisted Henry Gas Optimization Algorithm) model, the best characteristics from these retrieved features were selected. This is the best option for avoiding the dimensionality curse. The selected EC + AHGOA method obtain good accuracy (∼0.95) for tuning percentage 70 in pneumonia diagnosis from Chest X-ray images than some other previous techniques includes EC + AOA (∼0.92), EC + HGSO (∼0.93), EC + HGS (∼0.88), EC + PRO (∼0.90), and EC + BES (∼0.89).