NIS3D: A Completely Annotated Benchmark for Dense 3D Nuclei Image Segmentation

3D segmentation of nuclei images is a fundamental task for many biological studies. 1 Despite the rapid advances of large-volume 3D imaging acquisition methods and 2 the emergence of sophisticated algorithms to segment the nuclei in recent years, 3 a benchmark with all cells completely annotated is still missing, making it hard 4 to accurately assess and further improve the performance of the algorithms. The 5 existing nuclei segmentation benchmarks either worked on 2D only or annotated 6 a small number of 3D cells, perhaps due to the high cost of 3D annotation for 7 large-scale data. To fulfill the critical need, we constructed NIS3D, a 3D, high 8 cell density, large-volume, and completely annotated Nuclei Image Segmentation 9 benchmark, assisted by our newly designed semi-automatic annotation software. 10 NIS3D provides more than 22,000 cells across multiple most-used species in this 11 area. Each cell is labeled by three independent annotators, so we can measure the 12 variability of each annotation. A confidence score is computed for each cell, allow-13 ing more nuanced testing and performance comparison. A comprehensive review 14 on the methods of segmenting 3D dense nuclei was conducted. The benchmark was 15 used to evaluate the performance of several selected state-of-the-art segmentation 16 algorithms. The best of current methods is still far away from human-level accuracy, 17 corroborating the necessity of generating such a benchmark. The testing results 18 also demonstrated the strength and weakness of each method and pointed out the 19 directions of further methodological development. The dataset can be downloaded 20 here https://