Classification of Chest Radiographs into Normal and Pneumonia Class Using Fused Deep-Features

In human physiology, lungs are chief part of the respiration system and infection in lung is a medical emergency. Pneumonia is one of the lung infections, which severely affects the respiration system and may cause severe illness in children and elderly. The early diagnosis of the pneumonia is essential and the clinical level assessment of pneumonia's severity will be assessed clinically using chest X-Ray. This research aims in developing a deep-learning supported tool for pneumonia (viral+bacterial) detection. The phases of this tool includes; data collection, resizing and contrast enhancement, feature extraction with pre-trained deep-learning (PD) method, deep-features reduction with 50% dropout and serial features integration, and binary classification and verification. The experimental investigation is performed on the chosen database using; (i) individual deep-feature, and (ii) reduced-features and (iii) fused deep-features. During this investigation, Random-Forest (RF) assisted classification is executed and this process provided an accuracy of >98% when the fused deep-features are considered. In the future, the performance of this scheme can be verified using the Covid19 database found in the literature.