Advanced and Minor Lung Disease Severity Classification Using Deep Features

• Main Authors: Chia Ming, Joel Than, Noor, Norliza Mohd, Rijal, Omar Mohd, Kassim, Rosminah M., Yunus, Ashari
• Format: Conference Proceeding
• Language: English
• Subjects: Classification; Deep; Disease; Lung; Severity
• ISSN: 2642-6471
• DOI: 10.1109/ICSIPA45851.2019.8977774

A key component of Computer Aided Diagnosis (CADx) systems is classification. This stage involves feature extraction. Deep features have shown to be an emerging area of research in various fields including medical imaging. Most recent deep works in the lung area focused using Convolution Neural Networks (CNN). However, these works have drawbacks such as over-classifying and are not reflective of the real world. Therefore this study's aim is to develop a lung disease severity classification (LDSC) system. Information on lung disease severity can be beneficial to clinicians to plan treatments. This study used high resolution computed tomography (HRCT) scans from Hospital Kuala Lumpur where there were 81 diseased (Interstitial Lung Disease) and 15 normal patients. Deep features were obtained using a pre-trained network (Alexnet). This studied used four classifiers Support Vector Machine (SVM), Naïve Bayes, Random Forest and K-NN. A high accuracy of 100% was achieved using SVM.