A brain tumor MR image segmentation algorithm based on logical stick-breaking and Markov random field

Accurate segmentation of brain tumor MR images is of great significance for clinical diagnosis and treatment. However, existing methods still face obvious limitations such as computational complexity and the need to preset the number of clusters under the influence of noise interference and tumor heterogeneity. This paper innovatively proposes an adaptive segmentation algorithm that integrates a Logical Stick-Breaking mechanism with a Markov random field. By establishing a probabilistic association model between pixel spatial positions and cluster labels and introducing neighborhood consistency constraints, while adopting a Variational Bayesian(VB) inference framework to ensure the algorithm's efficiency and stability, it successfully addresses the problems of blurred segmentation boundaries and dependence on manually preset cluster numbers in complex noise environments in traditional methods. Experimental verification shows that this method significantly improves the automatic recognition accuracy of multiple brain tumors while maintaining high computational efficiency, especially with a higher detection rate for small lesions compared to existing mainstream algorithms, providing a more reliable quantitative analysis tool for clinical diagnosis.