Search

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. [ABSTRACT FROM AUTHOR]

Quantitative analysis in mammography is commonly carried out using a customized breast phantom. Noise reduction algorithms such as Wiener, median, and median modified Wiener filter (MMWF) with 3 × 3, 5 × 5, 7 × 7, and 9 × 9 mask sizes are applied to digital medical images because the digital images contain noise distributions such as Gaussian noise. The purpose of this study was to validate a customized breast phantom consisting of three positions for the inner structure (Condition A: 0.5 cm; condition B: 2.0 cm, condition C: 3.0 cm from the top surface) by applying the noise reduction algorithms listed above. The acquired images were evaluated using signal-to-noise ratio (SNR), contrast-to-noise ratio (CNR), and normalized noise power spectrum (NNPS) parameters. When comparing conditions A, B and C, the SNR results using condition A were 1.2 and 1.3 times better, the CNR results were 1.9 and 3.3 times better, and the NNPS results were 1.2 and 1.3 times better than for conditions B and C, respectively. The application of algorithms was useful for the reduction of noise distribution; in particular, the MMWF algorithm with a 9 × 9 mask size presented the best performance in terms of SNR, CNR, and NNPS results. In conclusion, we confirmed that the location of the breast lesion can affect the image quality under the same scan conditions and that noise reduction algorithms are essential processes to improve image quality. [ABSTRACT FROM AUTHOR]