Chest X-Ray Patch Classification for Tuberculosis Detection

Tuberculosis (TB) is one of the major global health issues, especially in developing countries. Although newly diagnosed TB patients can recover with a high cure rate, there are still many TB patients dying due to late diagnosis. A common way to diagnose TB is through an examination of the chest X-ray image (CXR) of the patient. Due to its high sensitivity, this method is preferred to others, and is vital for screening and triage for TB patients. Unfortunately, this examination requires an experienced radiologist to read the CXR image. In developing countries like Indonesia, the number of radiologists is still lacking and not well distributed across regions. In order to expedite the diagnosis, an automated system for detecting TB from CXR images may be able to help physicians examine more CXR images faster. In this paper, we proposed texture analysis of CXR images of Indonesian patients. The images were divided into local patches and represented using a variety of feature combinations. We then attempted to classify these patches into as normal or TB lesion patches using Support Vector Machine (SVM). Our results show that the combination of Principal Component Analysis (PCA) with GLCM and Hogeweg texture features obtain the best overall results compared to the baseline, with an accuracy of 91.2%, sensitivity of 97.1%. and snecificity of 87.2%.