Transfer Learning based Detection of Pneumonia from Chest X-Ray Images

Abstract

Pneumonia is an inflammatory condition affecting the small air sacs known as the alveoli present in the lungs. Despite the availability of vaccines for certain types it is known to be one of the leading causes of death across all age groups around the world. Chest X-Ray (CXR) images, blood test or sputum culture are standard techniques primarily used by doctors to confirm their diagnosis but is prone to human error due to huge imbalance between number of potential patients and doctors. Deep learning based computer aided technology with reasonably good accuracy and precision can aid the doctors by eliminating the benign cases. In this paper, a transfer learning based convolutional neural network (CNN) architectures is proposed for classifying CXR images into healthy and pneumonia affected with high accuracy and precision. The proposed method uses three different transfer learning architectures, viz. VGG - 16, VGG - 19 and Inception Resnet V2 for comparison and is found to provide best results with VGG - 19 architecture. An accuracy of 95.82% with 98.55% precision, 96.20% specificity and 95.67% sensitivity are obtained with the help of VGG-19 which is superior to any existing solution known to the authors.