Chest infection classification from X-ray images using enhanced multisource transfer learning with voting system

Abstract

Chest infection is a major health threat in most regions of the world. It is claimed to be one of the top causes of postoperative death after fragility hip fractures, according to a study presented in 2011. With the invention of deep learning in machine learning, implementation in Computer Aided Diagnosis system which utilizes deep neural networks for learning, classification, generation and even clustering has allowed X-ray image classification to be more accurate. The improvement in medical image classification using transfer learning is further studied. In this thesis, a novel deep neural network model which is composed of two Convolutional Neural Networks (CNNs) with different depth of weight layers, where the prediction probabilities for all CNNs are fused to the voting system for chest X-ray image classification is proposed and presented. The performance and accuracy of several existing deep learning model are investigated and compared to the proposed model. The outcome of this work, we successfully classified chest infection in chest X-ray images using the proposed model with overall accuracy of 83.69%.