神经元分割性能。(a)在小鼠皮层深度100µm的MesoLF记录的2D切片中,比较MesoLF与CNMF-E(模板匹配和基于形状的选择步骤)的分割性能。绿圈:与地面事实强烈匹配的部分。蓝色圆圈:只出现在基本事实中的片段。洋红色圆圈:与基本事实不一致的部分。(b) CNMF-E(模板匹配和基于形状的选择步骤)和MesoLF分割中神经元检测性能的精度、灵敏度和f1评分的比较。与主图3h中的数据相同,为方便起见在此复制。横条高度:平均值。误差条:s.d。黑圈:n = 5次模拟运行。(c)顶部面板:包含神经元并显示散射的3D体积插图,用于图中其余部分的体积分割比较。 Schematic illustration of segmentation pipelines in CNMF-E (middle box) and MesoLF (bottom box). (d) 3D rendering of segmentation results from CNMF-E (left) and MesoLF (right). Magenta: Ground-truth neurons, green: segments. (e) Zooms into areas indicated by dashed rectangles in (d). (f) Comparison of the spatial similarity index of neurons paired between ground truth and output of CNMF-E (template matching and shape-based selection steps) versus MesoLF segmentation. p = 2.0e-9, paired one-sided Wilcoxon signed rank test. n = 63 neuron pairs. ** p < 0.01. (g) Histogram of spatial similarity indices of segmented neurons compared to ground truth by both methods (same data as in (f)). Credit:自然方法(2023)。DOI: 10.1038 / s41592 - 023 - 01789 - z