Fig. 1: Workflow for standardized implementation and benchmarking of nuclear segmentation algorithms on our curated in-house and external paired DAPI and annotated ground truth nuclei datasets. | Communications Biology

Fig. 1: Workflow for standardized implementation and benchmarking of nuclear segmentation algorithms on our curated in-house and external paired DAPI and annotated ground truth nuclei datasets.

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

Fig. 1

A Nuclear segmentation is performed on spectrally unmixed DAPI signal. Two datasets were used in this study- in-house imaged DAPI ROIs (vi) for which binary ground truth nuclei masks were annotated (ii–v) and an external, publicly available dataset which was merged with our in-house dataset (vii) for evaluation of parameter-free deep learning algorithms with enhanced statistical power. B Deep learning algorithms (implemented in Python) and classical algorithms (implemented in various platforms with a GUI) were evaluated. C ROIs were sampled from the fields (which were sampled from WSIs). The segmentation predictions of the different algorithms on the sampled ROIs were compared with the ground truth nuclear masks to compute object-level quantitative metrics for comparison and benchmarking of nuclear segmentation performance. Scalebar in red A(i), C 100 μm, and A(ii–v) 10 μm.

Back to article page