Fig. 2

MATISSE segmentation promotes both cell identification quantity and quality. a Numbers of cells were quantified using IMC and MATISSE segmentation methods for all analyzed regions of interest (ROIs). Lines link the datapoints per ROI. Paired t test was performed to test for significance. ****p < 0.0001. N = 45 images. b, c Overlap between manual annotations and predictions was quantified by recall score and b compared for MATISSE and IMC at varying intersection-over-union (IOU) thresholds, c displayed per ROI at IOU 0.6 and higher, lines link datapoints per ROI. Paired t test was performed to test for significance. ****p < 0.0001. N = 30 images. d Representative image of IOU values indicated by a color-scale labeling of the annotated events (red lining) that overlap with predictions by IMC or MATISSE. Black lines indicate outlines of the predictions. Scale bar 25 μm. e Fraction of split annotated events were quantified using IMC and MATISSE segmentation methods for all ROIs, lines link the datapoints per ROI. Paired t test was performed to test for significance. ****p < 0.0001. N = 30 images. f Edge intersection score per ROI was determined by quantifying intersection of predicted cell outlines by both methods with manually annotated nuclei, where a lower score corresponds to less overlap. Lines link the datapoints per ROI. Paired t-test was performed to test for significance. ****p < 0.0001. N = 30 images