Fig. 3: Comparison of nuclear segmentation performance of the pre-trained models of Mesmer, Cellpose (“nuclei” model), and StarDist (“2D_versatile_fluo” model) on the merged dataset using quantitative metrics. | Communications Biology

Fig. 3: Comparison of nuclear segmentation performance of the pre-trained models of Mesmer, Cellpose (“nuclei” model), and StarDist (“2D_versatile_fluo” model) on the merged dataset using quantitative metrics.

From: Quantitative benchmarking of nuclear segmentation algorithms in multiplexed immunofluorescence imaging for translational studies

Fig. 3

Mean (averaged across 60 ROIs) with error bars showing 95% confidence interval of A F1-score, D precision, E recall, F Jaccard index are evaluated at an IoU threshold of 0.5. The IoU threshold of 0.5 is the most lenient threshold required to ensure a maximum of one true positive predicted nucleus for each ground truth nucleus. B Boxes show the median and quartiles of the F1-scores (evaluated at 0.5 IoU threshold) across the 60 evaluation ROIs and whiskers extend up to 1.5 times the inter-quartile range with outliers shown. C Mean F1-score (averaged across 60 ROIs) at varying IoU thresholds, with area under the curve shown. A higher IoU threshold results in a stricter condition for classifying a predicted nucleus as a true positive. G Nuclear prediction computation time for each algorithm using CPU and GPU. Boxes show the median and quartiles of the computation time (n = 27 fields) and whiskers extend up to 1.5 times the inter-quartile range with outliers shown.

Back to article page