Denoising of magnetic resonance imaging using Bayes shrinkage based fused wavelet transform and autoencoder based deep learning approach

Denoising of medical images is of great concern as it plays a significant role in performance of computer aided diagnosis (CAD) systems. In real life scenarios, various conditions like vibration of magnetic coils due to rapid pulses of electricity contribute to noise during the procurement of medical images such as magnetic resonance imaging (MRI), computed tomography (CT), and ultrasound. The use of imaging modality depends on the type of disease and its severity, and MRI is the commonly employed imaging modality for diagnosis of second most common dreadful cancers in men known as prostate cancer. However, MRI is prone to certain noises as Gaussian and Rician making denoising one of the important steps in the CAD system. Traditional approaches used for denoising of MRI were prone to certain issues such as loss of data due to compression and preservation of edge details. Hence, this paper presents Bayes shrinkage based fused wavelet transform (BSbFWT) and Block based autoencoder network (BBAuto-Net) for removal of noise from MRI. Further, the performance analysis of the denoising approaches are performed using different metrics. Thus, the values of peak signal to noise ratio (PSNR), mean squared error (MSE), structural similarity index metric (SSIM) and mean absolute error (MAE) for proposed BB-Autonet is found to be 28.029, 89.354, 0.581 and 21.802 for combined Gaussian and Rician noise. Whereas, the values of PSNR, MSE, SSIM and MAE for proposed BSbFWT are found to be 29.028, 81.33, 0.747 and 21.962 for combined Gaussian and Rician noise.

Automated summary

Denoising of medical images is crucial for the performance of computer aided diagnosis (CAD) systems, with MRI being susceptible to Gaussian and Rician noises. The proposed methods, BSbFWT and BBAuto-Net, show promising results in removing noise from MRI with performance evaluations using various metrics.