Didik Bani Unggul, Nur Iriawan, Irhamah Irhamah, Miftah Fahira, Ulfa Siti Nuraini
Mixture models are probabilistic frameworks that represent data through a combination of statistical distributions, termed component distributions. In image segmentation, mixture models will allocate pixels to components by analyzing feature similarities, with each component defining a distinct segmented region. This approach enables automated and accurate identification of regions of interest based on statistical coherence. However, review articles on the development and methodological variations of mixture models are still lacking, especially in narrow domains. To address this issue, we present a systematic review that specifically focuses on the application of mixture models to biomedical image segmentation. After going through the search and selection process, we collected 110 relevant articles whose key information were analyzed in more depth, ranging from the variation of biomedical cases to the used mixture modeling methodology. The results show that the key strength of mixture modeling is its flexibility in many factors, from theoretical foundations to practical case studies. We observed wide variation in aspects such as the choice of probability distributions, how the number of components is determined, and integration with other methods to enhance performance. Finally, we highlight the limitations and challenges of this approach, outlining key focuses for future research to improve its applicability. © 2013 IEEE.
Institut Teknologi Sepuluh Nopember, Department of Statistics, Surabaya, 60111, Indonesia; Universitas Negeri Surabaya, Data Science Department, Surabaya, 60213, Indonesia
