Magnetic Resonance Image Denoising using Patchwise Convolutional Neural Networks
Ram Bhawan Singh, Lakhwinder Kaur · International Conference on Computing for Sustainable Global Development · 2021
Magnetic Resonance Image (MRI) signal is usually degraded with additive white Gaussian noise in its real and imaginary components. The spatial 2D and 3D digital images are reconstructed from the complex domain noise corrupted MRI signal. The resulted noisy image follows the Rician probability distribution in which noise become signal-dependent. To provide quality MRI for clinical applications, these images are required to pre-process automatically by computerized algorithms. Thus, it becomes necessary to recover noise-free MRI from its noise degraded counterpart for accurate automatic analysis. In this paper, machine learning convolutional neural network-based (CNN) MRI denoising and artifacts removal technique is presented. The noise filtering networks learn the image details from the image patches pixel-by-pixel from noise residuals to restore the detailed image features in an end-to-end feed-back approach. The proposed technique shows comparable performance with existing other image restoration techniques without losing important image information.