Objective In vivo Micro-PET/CT imaging of mouse/rat brain has been widely used to non-invasively monitor brain and provides researchers a better understanding of therapeutic effects in models of human neurological disease. For the need of further processing, extraction of brain tissue from head is required and vital. Methods An automatic multi-step combination methods was proposed based on an improved level set framework, which includes (1) Use Fuzzy-C-Means method together with threshold and morphology methods to get the initial level-set surface automatically. (2] Combine gradient vector flow to enhance the gradient contrast and enforce the surface move toward to the object's surface much faster, especially obtain a significantly improvement in the regions of forehead and the joint between brain and neck. (3) Introduce an automatic stop condition based on average bandwidth energy maximization to overcome the leakage problem. Results 3 Micro-CT images of rat and 3 of mouse have been tested using the proposed methods and the average accuracy has increased by 33% for rat and 6.7% for mouse. The average processing duration for rat and mouse are about 8 minutes and 4 minutes, respectively. Conclusions The proposed methods were proved that it can be effectively used for Micro-PET/CT imaging of mouse/rat brain segmentation and have a great improvement on accuracy and convenience,%目的 提出一种基于改进水平集方法,能对Micro-CT 鼠脑图像进行有效分割的自动化方法.方法 首先利用模糊C均值聚类方法,并结合阈值分割和形态学方法自动设置水平集三维初始表面轮廓;其次使用梯度矢量流增强水平集在窄细等复杂结构处的演化能力;最后提出一种判断演化表面轮廓附近平均带宽能量是否达到最大化来设定停止条件;最终自动化获得准确分割的结果.采用基于区域法的评价方法,计算两个对应区域的重叠比例来验证该方法的改善效果.结果 利用以上方法分别对3例大鼠图像和3例小鼠图像进行测试,以重叠比例来衡量准确率分别在88.3% 和87.4%,与已有的经典方法相比,平均准确度分别提高了33% 和6.7%.平均处理时间分别为8min和4min.结论 本方法被证明能有效应用于Micro-CT 图像中鼠脑组织的分割和提取,在准确性和便捷性上都有较大改进.
展开▼
