MR medical image enhancement: an integration of residual approximation and contrast enhancement approach

Abstract

Magnetic Resonance Images are mainly corrupted by Gaussian and Rician noises during their acquisition, which degrades the calibre of post-processing diagnostics applied to MR data, such as segmentation, registration, morphometry etc. A pre-processing technique such as MR image enhancement is needed for precise diagnostic results. Recently, deep learning techniques are gaining much popularity in various biomedical applications due to their accuracy when trained with a huge volume of biomedical images. This article presents a deep learning-based pre-processing mechanism which integrates the residual approximation and contrast enhancement method. The denoised image is estimated using a denoising Convolutional Neural Network (CNN) and residual approximation. The contrast of the denoised image is enhanced using histogram equalization. To analyze the quality of enhanced image, metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index Measure (SSIM) and Mean Squared Error (MSE) are considered. The experimental results on synthetic and clinical data show that the proposed method outperforms the existing methods in terms of PSNR, SSIM and MSE. The proposed method obtained a PSNR of 40dB, SSIM of 99%, and MSE of. 0052 for Gaussian noise addition and for Rician noise addition, attained a PSNR of 38 dB SSIM of 99% and MSE of. 0084 for σ = 9 %.

More information

Identifiers

ORCID for Dr. Divya Saleela: orcid.org/0000-0002-7302-7146

Catalogue record

Export record

Altmetrics