Classification of Pneumonia in Chest Radiographs using Deep Learning Techniques

Pneumonia is a respiratory illness due to bacterial and viral infection prevailed in many people and frequently occur in developing/underdeveloped countries with lack of purification, hygiene, proper medical facilities in crowded areas. The paper establishes the fact that there are accounts of patients ending up with a condition of pleural effusions, this happens when the fluid accumulates in the lungs leading to respiratory problems. This is because knowing that a child or even an adult has pneumonia should prompt the medical officers to commence appropriate curative measures with the intention of raising the survival rate. Chest X-ray is also most commonly used in diagnosing pneumonia. However, in the case of evaluation of chest X-rays several points are quite ambiguous and it involves variability to certain extent. Our approach is about developing a CAD system that can assist in the classification of the pneumonia from chest X-ray images. Due to high adaptation of few available data, we deployed various deep transfer learning approaches which involve three approaches: MobileNet, MobileNet + LSTM, and EfficientNetB0. The previous approach with ResNet50 model did not reach high accuracy. Therefore, to improve the performance of the proposed method, we dispose the models of transfer learning. The above mentioned approach was experimented with pneumonia X-ray images dataset which is public domain. It is used to predict the outcomes in images as Normal, Pneumonia-Bacterial or Pneumonia-Viral.