Pneumonia Detection in chest X-rays: a deep learning approach based on ensemble RetinaNet and Mask R-CNN

Abstract

Pneumonia is a bacteria, virus, or fungus that infects one or both lungs, causing the alveoli to fill with fluid or pus. More than 15% of deaths include children under age five are caused by pneumonia globally. This disease is usually diagnosed by chest X-ray. The rich labeled data sets verify the effectiveness of deep learning technology. In this study, this paper describes a method based on deep learning to automatically identify and locate the location of pneumonia on chest X- ray images. We have constructed a new pneumonia detection model, which is a new pneumonia detection model obtained by the ensemble of the improved RetinaNet and Mask R-CNN models. Among them, the improved RetinaNet model is ensemble of the RetinaNet models under different backbone networks ResNet-50 and ResNet-101. Similarly, we also use different backbone networks ResNet-50 and ResNet-101 for the Mask R-CNN network. The improved Mask R -CNN model is obtained by ensemble pneumonia detection models under different backbone networks. Finally, ensemble improved RetinaNet and Mask R-CNN pneumonia detection models. This paper validated our method on the dataset of 26,684 chest radiographs published on Kaggle, and achieved a recall of 0.813 and a mAP of 0.2283. Our approach achieves robustness through key modifications to the training process and novel processing steps that incorporate multiple models. A good performance evaluation was obtained.