An application of faster R-CNN for chest X-ray digital image classification / Mohd Taufik Rahmat

Non-digitalized chest X-ray is an effective, low-cost screening tool, and it is important to indicate pathologies. However, there are some cases of misinterpretation in the diagnostic process. Reading and interpret chest X-ray may be a simple task for a radiologist, but not every doctor can do it the same. This paper aims to evaluate the performance of chest X-ray image classification using Faster R-CNN architecture. To develop a chest x-ray classifier model, Tensorflow package was used with python. The results show the propose model performance accuracy is 62%. The model then was compared to random selected one medical student and general practitioner. The model shows better in term of performance to classify chest x-ray images with 62% accuracy compared to selected medical students and general practitioners with their accuracy score of 56% and 50% respectively. In term of chest X-ray interpretation in this study, the result shows that the model performance is more reliable to use for chest x-ray images classification. Tough the model performance is better, but in medical field reality, it is still far from the standard to be applied. With 62% accuracy, the model is unsafe to use. The future works are to gain more knowledge from radiologist expert to improve chest -x-ray classifier performance.