Denoising Autoencoder for the Removal of Noise in Brain MR Images

Abstract

Medical imaging methods like X-rays, ultrasound, Computed Tomography (CT) scans and Magnetic Resonance Imaging (MRI) can show structures of the internal body in great detail. However, there is unavoidable and ubiquitous noise in images created by medical imaging equipment. The use of noise reduction methods in medical imaging has therefore become crucial. In this article, we are focusing primarily on eliminating noise from MRI images of the brain. In terms of revealing details about the location and size of tumors, MRI is quite effective. Here we are proposing a Convolutional Denoising Autoencoder for removing noise from these images. Convolutional autoencoders can gently extract the data while preserving the spatial information of the image data. As a result, we can denoise with greater accuracy while using less computation and data. Our model got an SSIM value of 0.85 and a PSNR value of 30 dB.