The Effect of Impulse Noise on Accuracy of Clustering of Chest Radiographic Images in Pneumonia

The effect of salt-and-pepper noise on the accuracy of clustering of lung X-ray images is considered. The analysis is performed for cases of healthy lungs and lungs with signs of bacterial and viral pneumonia. Fifty images of different sizes are considered for each of the three classes. The level of salt-and-pepper noise is taken from 1 % to 5%. The number of clusters for each image is chosen to be three (elbow method). The differences between clusters of noisy and original images are studied by the IoU metric. It is concluded that images with viral pneumonia have a better match for this metric than the other two classes. In other words, noise has a lesser effect on the clustering results for viral pneumonia. Cases with a number of clusters greater than three are also analyzed.