A Deep Learning Method for Pneumonia Detection Based on Fuzzy Non-Maximum Suppression

Pneumonia is one of the largest causes of death in the world. Deep learning techniques can assist doctors to detect the areas of pneumonia in the chest X-rays images. However, existing methods lack sufficient consideration for the large variation scale and the blurred boundary of pneumonia area. Here, we present a deep learning method based on RetinaNet for pneumonia detection, by introducing the multi-scale feature extract network Res2Net and improving non-maximum suppression (NMS) algorithm. We proposed a novel NMS algorithm, named Fuzzy Non-Maximum Suppression (FNMS), by fusing the predicted boxes with high overlap scores to get a more robust predicted box. We apply FNMS in the single model case and the model ensemble case. In the single model case, improved RetinaNet is obviously better than baseline. In the model ensemble case, the final predicted box fused by FNMS is better than three other model ensemble methods NMS, Soft-NMS, and weight boxes fusion. Experimental results on pneumonia detection dataset verify the superiority of the FNMS algorithm.