Performance evaluation of linear and nonlinear filters for despeckling B mode foetal heart ultrasound images

Abstract

ABSTRACT Early detection of congenital heart disease (CHD), one of the most commonly occurring congenital defects, is important to reduce mortality rates. The major drawback of ultrasound imaging is the inherent speckle noise, making visual examination of anatomical structures a challenging task. This study discusses the effect of denoising using different linear and nonlinear filters on B mode foetal cardiac ultrasound images. The exhaustive study of the performance of linear filters such as mean, Laplacian, and Wiener, and nonlinear filters such as median, anisotropic diffusion, non-local means, and bilateral filters has been carried out. Performance of the filtering technique is evaluated using various fidelity measures and a new assessment parameter edge preserving index (EPI) is also used for evaluation. Non-local means filter outperforms other filtering techniques with average fidelity values as follows: PSNR: 67.97 dB, MSE: 0.01283, RMSE: 0.109, average difference (AD): 0.005, normalized absolute error (NAE): 2 × 10−4, mean absolute error (MAE): 0.011, EPI: 0.999. This work serves as a reference on pre-processing techniques that can be applied to foetal cardiac ultrasound images for beginners.