Cheng Zhaoning, Song Zhijian
Digital Medical Research Center, Fudan University, Shanghai 200032, China.
Sheng Wu Yi Xue Gong Cheng Xue Za Zhi. 2013 Apr;30(2):265-71.
Magnetic resonance imaging (MRI) is usually obscure and non-uniform in gray, and the tumors inside are poorly circumscribed, hence the automatic tumor segmentation in MRI is very difficult. Region-scalable fitting (RSF) energy model is a new segmentation approach for some uneven grayscale images. However, the level set formulation (LSF) of RSF model is not suitable for the environment with different grey level distribution inside and outside the intial contour, and the complex intensity environment of MRI always makes it hard to get ideal segmentation results. Therefore, we improved the model by a new LSF and combined it with the mean shift method, which can be helpful for tumor segmentation and has better convergence and target direction. The proposed method has been utilized in a series of studies for real MRI images, and the results showed that it could realize fast, accurate and robust segmentations for brain tumors in MRI, which has great clinical significance.
磁共振成像（MRI）的灰度通常模糊且不均匀，内部肿瘤边界不清，因此MRI中的肿瘤自动分割非常困难。区域可缩放拟合（RSF）能量模型是一种针对某些灰度不均匀图像的新分割方法。然而，RSF模型的水平集公式（LSF）不适用于初始轮廓内外灰度分布不同的环境，而MRI复杂的强度环境总是使其难以获得理想的分割结果。因此，我们通过一种新的LSF对模型进行了改进，并将其与均值漂移方法相结合，这有助于肿瘤分割，且具有更好的收敛性和目标方向。所提出的方法已应用于一系列真实MRI图像的研究中，结果表明它可以实现对MRI中脑肿瘤的快速、准确和稳健分割，具有重要的临床意义。
