DNN-driven hybrid denoising: advancements in speckle noise reduction

Image denoising plays a crucial role as a preliminary step in medical image analysis, facilitating accurate illness diagnosis and therapy in the context of ultrasound images. Ultrasound is a significant part of medical imaging. It is one of the most used medical diagnostics techniques. Ultrasound images frequently experience degradation due to the presence of various types of noise during the process of capture or transmission, leading to a reduction in contrast and a deterioration in overall image quality. Speckle noise is a type of noise that caused degradation of ultrasound pictures. The inclusion of speckles inside a picture might potentially impede the precise representation of intricate elements and lower the contrast within delicate tissues, leading to a reduction in visual fidelity. So a hybrid denoising algorithm is implemented to enhance the quality and contrast of ultrasound images. The hybrid denoising algorithm consists of a combination of a modified perona-malik model and deep learning. This study aims to examine the outcomes of a hybrid model by using several denoising methods with and without the incorporation of a deep learning model for ultrasound pictures. The denoising techniques that have been implemented include interpolation, adaptive median, wavelet, anisodff-2D, and adaptive gue. Multiple assessment metrics are employed to assess the effectiveness of the utilized approaches, including PSNR, SSIM, NIQUE, mean square error, and edge intensity (Edge I). The model undergoes testing on many categories of ultrasound images, encompassing both normal and malignant instances, while considering the noise variance values of [20, 50, 80, 100].