Diagnosis Model in Smear-Negative Pulmonary Tuberculosis Using Faster R-CNN

Abstract

Background: One of the most important human organs in the respiratory system is the lung. The main function of the lung is the respiration process, which is responsible for pumping air into the body. The health of the lung organs is very important, because if this organ is disturbed it will affect the health of the rest of the body. One of the diseases that attacks the lungs is Tuberculosis (TB). TB disease can be cured, but if it is delayed in getting treatment it can increase the risk of death. Method: This research developed a Smear Negative Pulmonary Tuberculosis diagnosis model using the Deep Learning method using the Faster R-CNN algorithm. The data used in this research are x-ray images of the lungs at the Jakarta Repository Center - Indonesian Tuberculosis Eradication Center (JRC-PPTI) clinic, totaling 220 datasets. At the preprocessing stage, the images used for training and testing were used with a size of 1280 x 1280 to see the effect on the accuracy of the prediction results of the Faster RCNN model. The test results are in the form of accuracy values that reflect the performance of the Faster RCNN model in classifying normal (without TB) and abnormal (with TB) test data. Results: The research implementation carried out the training process and testing process for 75% of training images, and 25% for testing images. Training images are labeled using the Img label. In the testing stage of the faster RCNN model, the accuracy value was 62.04%, precision was 40.00%, recall was 64.52% and F1-score was 49.38%. Conclusion: From the results of this research it is concluded that the Faster RCNN model test results using the ResNet 50 model have an accuracy value of 62.04%, Precision of 40.00%, Recall of 64.52% and F1-score of 49.38%.