Fig. 2: CellSPA performance evaluation framework.

a Schematic showing the cell segmentation evaluation framework with five complementary categories. b Bar plots showing overall characteristics, including the number of cells [left], and the number of transcripts [right] for each of the 11 methods. c Boxplots of cell-level quality metrics with total number of transcripts [left] and total number of genes [right]. The number points for each box includes the number of cells detected by each method (N = Chromium: 22,294; Cellpose (nuclei): 99,693; BIDCell: 103,209; 10x (nuclei): 126,515; 10x: 160,254; JSTA: 107,131; Cellpose nuclei dilated: 104,307; Cellpose cell: 87,046; Voronoi: 106,227; Watershed: 105,527; Baysor: 177,437; Baysor (no prior): 191,698), and ranges from the first to third quartile with the median as the horizontal line. The boxplot’s lower whisker extends 1.5 times the interquartile range below the first quartile, while the upper whisker extends 1.5 times the interquartile range above the third quartile. d Gene-level quality metric represented by a scatter plot of the percentage of cells expressed for each gene in the segmented cells (y-axis) vs. the nuclei (x-axis). e Cell morphology metrics represented by the elongation values between the segmented cells (y-axis) and nuclei (x-axis), where each dot represents the average elongation for each cell type and the Pearson correlation between the elongation values of nuclei and segmented cells is noted in the top left corner. f Scatter plot between correlation the elongation values of nuclei and segmented cells (y-axis) and average total number of transcripts per cell (x-axis) based on average expression. Source data are provided as a Source Data file.