Ulfa Siti Nuraini, Nur Iriawan, Kartika Fithriasari, Irhamah, I. Ketut Alit Utamayasa
Due to increased noise, cardiac ultrasound imaging is challenging. Their variable quality, characterized by speckle noise, low spatial resolution, and acoustic dropouts, makes ultrasound segmentation a complex and demanding task. To reduce noise in ultrasound imaging, this research proposed a segmentation approach that enhances model-based clustering. The combination of a non-symmetrical data distribution (MSTBurr) with spatial statistical-mechanical effects (the Ising model) is the primary contribution of this article. The model was developed to handle the laxity of the Gaussian Mixture Model (GMM), which used symmetric distributions and ignored pixel neighbors. Hybrid of Variational Inference and Gibbs Sampling in the MCMC framework is used to estimate the model using the Bayesian technique. Using the G-Mean and IoU with GMM and Bayesian Gaussian Mixture, MSTBurr-Ising performance was assessed, exceeding 70%. This article also used simulated data with varying noise percentages compared to other methods. The segmentation results show that segmentation using the MSTBurr-Ising model reduces noise better than Gaussian-based models. When it comes to cardiac ultrasound image data and simulation image data with different types of noise, MSTBurr-Ising performs better than GMM and Bayesian Gaussian Mixture. The results of this modeling hopefully can help paramedics treat patients more accurately by improving their interpretation of cardiac ultrasound images. © 2025 IEEE.
Department of Statistics, Institut Teknologi Sepuluh Nopember, Department of Data Science, Universitas Negeri Surabaya, Surabaya, Indonesia; Department of Statistics, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia; Department of Child Health, Cardiology Division Dr. Soetomo General Academic Hospital, Surabaya, Indonesia
