Chest X-Ray Image Denoising Based on Convolutional Denoising Autoencoder

Abstract

Nowadays, medical imaging plays important role in medical settings to obtain high resolution images for the human body. The medical imaging techniques usually suffer from many types of noises such as gaussian, salt and pepper and speckle noises. So, getting a high-resolution body image is so difficult. The accurate medical images is necessary for diagnosis of many diseases. In this paper, medical imaging denoising technique based on convolutional denoising autoencoder is proposed. The NIH chest X-Ray dataset has been used for the training and testing of the proposed model. The model consists of 10 layers to learn the representation of the noise in the image and then reconstruct a new image without the noise. The model performance evaluated by using mean squared error and peak signal to noise ratio. For the training purpose we added gaussian noise to the dataset. The total number of images used is 25,000 splitted into training set 22,500 images and testing set 2500 images. The model achieved excellent results on the testing set with 0.01 mean squared error.