IA-Noise2Noise: an Image Alignment Strategy for Echocardiography Despeckling

Yuxuan Li,Wenkai Lu,Patrice Monkam,Yonghao Wang

DOI: https://doi.org/10.1109/ius51837.2023.10307810

2023-01-01

Abstract:Although echocardiography plays a crucial role in cardiovascular disease diagnosis, the presence of speckle noise in echocardiography images can hinder their diagnostic effectiveness. Recent studies have focused on using deep learning models to reduce speckle noise in echocardiography. However, these deep learning models’ performances tend to be limited on real data as they are usually trained with synthesized images due to the unavailability of real noise-free echocardiography images. Moreover, although the Noise2Noise algorithm which is widely adopted for natural image denoising can be used for ultrasound image despeckling without using clean images as training targets, this algorithm still requires noisy image pairs which are hard to acquire for echocardiography due to the cardiac motion. To address this issue, we propose the IA-Noise2Noise algorithm, which incorporates an image alignment strategy into the Noise2Noise algorithm. Experiment results indicate that our method outperforms previous methods in terms of despeckling performance.