Level set formulation for automatic medical image segmentation based on fuzzy clustering

• Published in: Signal processing. Image communication Vol. 87; p. 115907
• Main Authors: Yang, Yunyun, Wang, Ruofan, Feng, Chong
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.09.2020; Elsevier BV
• Subjects: Algorithms; Clustering; Dynamic constrained term; Energy conservation; Fuzzy clustering; Image segmentation; Level set method; Medical image segmentation; Medical imaging; split Bregman method; split Bregman method; Fuzzy clustering; Dynamic constrained term; Level set method; Medical image segmentation
• ISSN: 0923-5965; 1879-2677
• DOI: 10.1016/j.image.2020.115907

The level set method is widely used in medical image segmentation, in which the performance is seriously subject to the initialization and parameters configuration. An automatic segmentation method was proposed in this paper, which integrates fuzzy clustering with level set method through a dynamic constrained term in the new energy functional. It is able to use the results of fuzzy clustering directly, which can control the level set evolution. Moreover, the added constrained term is changing continuously until getting the final results. Such algorithm eliminates the manual operation a lot and leads to more robust segmentation results. With the split Bregman method, the minimization of the new energy functional is fast. The proposed algorithm was tested on some medical images and also compared with other level set models and the state-of-the-art method such as U-Net. The quantitative and qualitative experimental results show its effectiveness and obvious improvement for medical image segmentation. •An automatic medical image segmentation method based on fuzzy clustering was proposed in this paper.•A dynamic constrained term was added into the new energy functional, which can help us get more accurate results.•Quantitative results such as DICE values, precision, f-measure and computation times are presented.•Comparative results with other level set methods and the U-Net method show excellent performance of our model.•We also discussed the sensitivities of parameters in the new dynamic constrained term, which show that it is robust.