Ising Neo-Normal Model (INNM) for Segmentation of Cardiac Ultrasound Imaging

Ultrasound imaging is challenging due to more noise. This paper proposed a segmentation model that improves model-based clustering that can be used for reduced noise in ultrasound imaging. The main contribution is combining spatial statistical mechanic effects (the Ising model) with non-symmetrical data distribution (Neo-Normal), which is called the Ising Neo-Normal Model (INNM). The model was achieved due to the limitation of GMM (Gaussian Mixture Model), which used symmetrical distributional and ignored the neighbors of pixels. Model estimation used the Bayesian method with the MCMC framework that combines Variational Inference with Gibbs Sampling. The performance of INNM was evaluated using the G-Mean and Jaccard Index with GMM and Bayesian Gaussian Mixture, and it was over 70%. INNM also gives better results than GMM and Bayesian Gaussian Mixture in cardiac ultrasound image data and simulation image data with various kinds of noise.