Although conventional denoising filters have been developed for noise reduction from digital images, these filters simultaneously cause blurring in the images. To address this problem, we proposed the fast non-local means (FNLM) denoising algorithm which would preserve the edge information of objects better than conventional denoising filters. In this study, we obtained thoracic computed tomography (CT) images from a male adult mesh (MASH) phantom modeled by computer and a five-year-old phantom to perform both the simulation study and the practical study. Subsequently, the FNLM denoising algorithm and conventional denoising filters, such as the Gaussian, median, and Wiener filters, were applied to the MASH phantom image adding Gaussian noise with a standard deviation of 0.002 and practical CT images. Finally, the results were compared quantitatively in terms of the coefficient of variation (COV), contrast-to-noise ratio (CNR), peak signal-to-noise ratio (PSNR), and correlation coefficient (CC). The results showed that the FNLM denoising algorithm was more efficient than the conventional denoising filters. In conclusion, through the simulation study and the practical study, this study demonstrated the feasibility of the FNLM denoising algorithm for noise reduction from thoracic CT images.

Ultrasound imaging is widely used for accurate diagnosis due to its noninvasive nature and the absence of radiation exposure, which is achieved by controlling the scan frequency. In addition, Gaussian and speckle noises degrade image quality. To address this issue, filtering techniques are typically used in the spatial domain. Recently, deep learning models have been increasingly applied in the field of medical imaging. In this study, we evaluated the effectiveness of a convolutional neural network-based residual network (ResNet) deep learning model for noise reduction when Gaussian and speckle noises were present. We compared the results with those obtained from conventional filtering techniques. A dataset of 500 images was prepared, and Gaussian and speckle noises were added to create noisy input images. The dataset was divided into training, validation, and test sets in an 8:1:1 ratio. The ResNet deep learning model, comprising 16 residual blocks, was trained using optimized hyperparameters, including the learning rate, optimization function, and loss function. For quantitative analysis, we calculated the normalized noise power spectrum, peak signal-to-noise ratio, and root mean square error. Our findings showed that the ResNet deep learning model exhibited superior noise reduction performance to median, Wiener, and median-modified Wiener filter algorithms.

Recently, the total variation (TV) algorithm has been used for noise reduction distribution in degraded nuclear medicine images. To acquire positron emission tomography (PET) to correct the attenuation region in the PET/magnetic resonance (MR) system, the MR Dixon pulse sequence, which is based on controlled aliasing in parallel imaging, results from higher acceleration (CAIPI; MR-AC_Dixon-CAIPI) and generalized autocalibrating partially parallel acquisition (GRAPPA; MR-AC_Dixon-GRAPPA) algorithms are used. Therefore, this study aimed to evaluate the image performance of the TV noise reduction algorithm for PET/MR images using the Jaszczak phantom by injecting ¹⁸F radioisotopes with PET/MR, which is called mMR (Siemens, Germany), compared with conventional noise-reduction techniques such as Wiener and median filters. The contrast-to-noise (CNR) and coefficient of variation (COV) were used for quantitative analysis. Based on the results, PET images with the TV algorithm were improved by approximately 7.6% for CNR and decreased by approximately 20.0% for COV compared with conventional noise-reduction techniques. In particular, the image quality for the MR-AC_Dixon-CAIPI PET image was better than that of the MR-AC_Dixon-GRAPPA PET image. In conclusion, the TV noise-reduction algorithm is efficient for improving the PET image quality in PET/MR systems.