ALVLS: Adaptive local variances-Based levelset framework for medical images segmentation

Medical image segmentation is a very challenging task, not only because the intensity of the medical image itself is not uniform, but also it may be accompanied by the impact of noise. Although mathematics, computer science, medicine, and other interdisciplinary fields have begun to study the problem of medical image segmentation, and have put forward a variety of segmentation algorithms, there is still much room for further improvement and enhancement. In the process of medical image collection and reconstruction, it is easy to produce intensity inhomogeneity and noises, as well as interference from other tissues, resulting in the difficulty of accurate segmentation. In this paper, we propose the adaptive local variances-based level set (ALVLS) model to segment medical images with intensity inhomogeneity and noises, including cardiac MR images, brain MR images, and breast ultrasound images. According to the variance difference information, the ALVLS model can adjust the effect of the area term adaptively. The local intensity variances are designed to optimize the ability to resist noise, which improves the segmentation accuracy of medical images. We also propose the two-layer level set model for segmenting left ventricles and left epicardium simultaneously. Experimental results for medical images and synthetic images show the desirable performance of the ALVLS model in accuracy, efficiency, and robustness to noise. In medical image competition, the Dice coefficient is used to calculate the similarity between the segmentation result and the ground truth. Thus we do comparisons with other methods and show that the Dice coefficient of the proposed method is higher than other testing methods.