Medical Image De-Speckling Using Fusion of Diffusion-Based Filters And CNN

Abstract

Ultrasound images, although significantly low cost when compared to the likes of MRI scans and Biopsy, these images have the disadvantage of being laden with Speckle noise, which is a multiplicative noise, making it less accurate and competitive to the level of details provided on MRI scans. Improving the quality of ultrasound pictures so that they may be used as a viable alternative to MRI scans entails using a combination of different denoising filters to remove noise, based on their individual qualities. Following that, after the initial denoising, CNN was used to perform learning-based denoising to maximize the removal of speckle noise in the images. To simulate a real-life scenario, 750 ultrasound images were contaminated with speckle noise. After a comparative analysis of various methods, this paper's proposed model of combining filters and CNN model achieves a PSNR of approx. 52.25 dB on denoising a speckle- corrupted ultrasound. By combining the best of both worlds of Hybrid filters and CNN denoising, the proposed model achieves solid efficiencies for reducing speckle noise in Ultrasound images, making it a strong alternative.