Design on Lung TB Detection System Based on Statistical Features of X-ray Image Histogram

This paper presented the research result on the design of pulmonary TB (Tuberculosis) detection systems using a statistical approach. The study aimed to address two problems in detecting pulmonary TB by doctors, especially in remote areas of Indonesia, namely the long waiting time for patients to get the doctor's diagnosis and the doctor's subjectivity. We used hundreds of X-ray images from radiology department of Sardjito Hospital, Yogyakarta, as primary data and thirty data from various sources on the internet as secondary data. Using statistical approach, we exploited statistical image feature from image histogram, examined two statistical methods of PCA and LDA transformation for feature extraction, and two minimum distance classifier in image classification. We also used histogram equalization in the image enhancement process and bicubic interpolation in image segmentation and template making. Test results on primary and secondary data images show the identification accuracy of 94% and 83.3%, respectively.