iCut: an Integrative Cut Algorithm Enables Accurate Segmentation of Touching Cells.
iCut: an Integrative Cut Algorithm Enables Accurate Segmentation of Touching Cells.
复制标题
iCut：集成切割算法可准确分割接触单元
DOI：
10.1038/srep12089
复制
发表时间：
2015-07-14
影响因子：
4.6
通讯作者：
Chen S
中科院分区：
文献类型：
--
作者：
He Y;Gong H;Xiong B;Xu X;Li A;Jiang T;Sun Q;Wang S;Luo Q;Chen S
单个细胞在大脑的生物过程中发挥着重要作用。神经元的数量在正常发育和疾病进展过程中都会发生变化。高分辨率成像使直接计数细胞成为可能。然而，对接触细胞的自动和精确分割仍然是大规模和高度复杂数据集的主要挑战。因此，一个综合切割（iCut）算法，它结合了信息的空间位置和干预和凹轮廓与建立规范化的削减，已经开发。iCut涉及两个关键步骤：（1）首先用上述关于接触细胞的信息构造加权矩阵，以及（2）实施使用加权矩阵的归一化切割算法以将接触细胞分离成孤立的细胞。使用两种类型的数据对这种新算法进行了评估：开放的SIMCEP基准数据集和来自Nissl染色小鼠大脑的显微光学成像数据集。对于两个数据集，它分别实现了91.2 ± 2.1%/94.1 ± 1.8%和86.8 ± 4.1%/87.5 ± 5.7%的召回率/精确度。正如使用查全率和查准率的调和平均值量化的那样，iCut的准确率高于一些最先进的算法。这种全自动算法的更好性能可以有益于脑细胞结构的研究。
Individual cells play essential roles in the biological processes of the brain. The number of neurons changes during both normal development and disease progression. High-resolution imaging has made it possible to directly count cells. However, the automatic and precise segmentation of touching cells continues to be a major challenge for massive and highly complex datasets. Thus, an integrative cut (iCut) algorithm, which combines information regarding spatial location and intervening and concave contours with the established normalized cut, has been developed. iCut involves two key steps: (1) a weighting matrix is first constructed with the abovementioned information regarding the touching cells and (2) a normalized cut algorithm that uses the weighting matrix is implemented to separate the touching cells into isolated cells. This novel algorithm was evaluated using two types of data: the open SIMCEP benchmark dataset and our micro-optical imaging dataset from a Nissl-stained mouse brain. It has achieved a promising recall/precision of 91.2 ± 2.1%/94.1 ± 1.8% and 86.8 ± 4.1%/87.5 ± 5.7%, respectively, for the two datasets. As quantified using the harmonic mean of recall and precision, the accuracy of iCut is higher than that of some state-of-the-art algorithms. The better performance of this fully automated algorithm can benefit studies of brain cytoarchitecture.