Transfer Learning Based Model for Pneumonia Detection in Chest X-ray Images
Ola M. El Zein, Mona Soliman, A Elkholy, Neveen I. Ghali
Abstract
Pneumonia is a lung infectious disease caused by viral bacteria. It damages one or both lungs in humans. Expert radiotherapists must evaluate chest X-ray to detect pneumonia. As a result, designing an automated system for detecting pneumonia would be valuable for quickly treating the disease, especially in remote areas. This paper introduces a convolutional neural network-based model for reliably detecting pneumonic lungs from chest X-rays. This model can be used by doctors to treat pneumonia in the real world. This study proposes a hybrid model of EfficientNetB0 as a transfer learning-based model and support vector machine (SVM) hinge loss. The functionality of the pre-trained EfficientNetB0 model is used as feature-extractors followed by SVM classifier for the classification of abnormal and normal chest X-Rays. The statistical findings show that using a pre-trained EfficientNetB0 model and supervised classifier algorithm to evaluate chest X-ray images, specifically to detect Pneumonia, can be very beneficial. The proposed model achieves higher classification accuracy, precision, recall, and AUC values outperforming other state of art models with an overall accuracy of 97%.