Brain MRI Noise Reduction Using Convolutional Autoencoder

Abstract

AbstractThe Magnetic resonance imaging (MRI) machine will add random artifacts like intensity inhomogeneity, Gaussian noise, and Rician noise, while the image acquisition process. Due to hardware imperfection (magnetic fields, etc.), body motion during scanning, thermal noise, weak signal intensity (which causes low signal-to-noise ratio), etc. noises are present in the image. It is very difficult to diagnose of brain disorder if the MR image is corrected by Rician noise. In the past two decades, various algorithms have been proposed with different noise reduction performances. Recently machine learning and deep learning architectures outperform most of the conventional denoising algorithms. The Convolutional Neural Network (CNN) based residual learning architectures are showing promising performance in MRI noise reduction. The proposed algorithm makes use of convolutional neural network-based rician noise reduction using Augmented Autoencoder architecture, which increases the denoising performance with boosted MR image sample size. In particular, batch normalization and residual learning are applied to enhance noise reduction performance. To validate the proposed research work, tested two sets of MRI data from the Kaggle database. The proposed augmented autoencoder produced promising results at high rician noise levels and showed better performance over state of art deep learning architectures. The performance is measured in terms of Mean square error, Peak Signal to Noise Ratio, and Structural Similarity Index Measurements.KeywordsAugmented autoencoderCNNMRIResidual learningRician noise