Abstract

Pneumonia, a treatable illness, is a leading cause of death. It is crucial to quickly identify pneumonia in order to provide timely intervention. The use of automated diagnosis for pneumonia can greatly improve patient outcomes. However, developing accurate deep learning models for this purpose is challenging because of limited availability of well annotated training dataset. A new technique called deep supervised domain adaptation has been proposed for diagnosing pneumonia using chest x-ray images. The DSDA approach utilizes knowledge from a publicly accessible dataset called ChestX-ray14 and transfers it to a smaller, well-annotated dataset. DSDA guarantees efficient knowledge transfer by lining up the distributions of the source and target domains according to the underlying semantics of the training samples. Two subnetworks— one for the source domain and one for the target domain—are used in the process. Both sub-networks share feature extraction layers and undergo extensive training. This method, in contrast to standard domain adaptation methodologies, focuses on knowledge transfer from a source domain multi-label classification problem to a target domain binary classification task. The proposed technique is evaluated against various existing methods to assess its accuracy. The outcomes demonstrates DSDA method achieves promising performance in automating pneumonia diagnosis. In conclusion, this scholarly article introduces the DSDA method for automated pneumonia diagnosis using chest X-ray images. By effectively transferring knowledge, the proposed method shows potential in improving the accuracy of pneumonia diagnosis.