Object detection based approach for Automatic detection of Pneumonia

Pneumonia is a bacterial, viral, or fungal infection that causes swelling of the tissues in one or both lungs. Diagnosis for pneumonia requires an expert to find anomalies in the X-ray images, as it is a widespread disease, and it is not always possible for doctors to review all the cases. Most of the caused deaths are due to late diagnosis or some human error, which leads to an incorrect diagnosis of the disease. In this paper, we propose a novel approach to make an automated diagnosis and identify infected regions in the lungs. The proposed method uses a Mask RCNN object detector, which can identify anomalies in the lungs and also detect whether that anomaly is a pneumonia infection or is some other type of lung disease. The experimental results state that our proposed method gets an IoU score of 0.155, which is about 9% improvement other methods, i.e., YOLOv3, UNet, and Resnet used for Pneumonia infected tissue identification.