Pneumonia Detection on Chest X-Ray Using Convolutional Neural Network and Transfer Learning

Abstract

Diagnosis of thorax diseases is most commonly done by examining the Chest X-rays (CXR).
Computer-aided diagnosis (CAD) System assist radiologists in the interpretation of medical images.
CADimproves the quality of diagnosis and leverages the productivity of radiologists. Several Deep
Learning algorithms have been successfully implemented inorder to provide fully-automated, high
precision Computer-aided diagnosis (CAD) Systems. In this paper, a dedicated X-ray network trained
from scratch for Chest X-Ray image classification is explored. Further, transfer learning using
powerful network architecture like MobileNet, InceptionV3, VGG19 and ResNet-50 are investigated
indetail. Limited availability of annotated Chest X-ray images makes medical image classification
challenging. Transfer learning approach with andwithout fine-tuning can help overcome this issue by
transferring the knowledge gained from pre-trained networks to domain specific tasks.The
experimental study is tested onthe dataset of Chest X-Ray images, consisting of 2 categories,
Pneumonia and Normal.