NucFuseRank：核实例分割数据集融合与性能排序

Nuclei instance segmentation in hematoxylin and eosin (H&E)-stained images is pivotal for automated histological image analysis, with diverse applications in downstream tasks. While numerous machine learning and deep learning approaches have been proposed for nuclei instance segmentation, current research predominantly focuses on developing novel segmentation algorithms and benchmarking them on a limited number of arbitrarily selected public datasets. This practice often results in an incomplete and unsystematic comparison between methods, making it challenging to accurately assess the true performance of different algorithms across varied scenarios. To address this limitation, NucFuseRank introduces an innovative dataset fusion strategy, integrating multiple nuclei instance segmentation datasets from various sources to construct a larger, more diverse, and representative comprehensive dataset. This fused dataset aims to encompass the complexities of different tissue types, staining variations, image qualities, and nuclear morphologies, thereby providing a more robust foundation for evaluating algorithm generalizability. Building upon this, NucFuseRank further establishes a systematic performance evaluation framework and ranking mechanism. This framework not only considers traditional segmentation metrics (e.g., Dice coefficient, Jaccard index, average precision) but also incorporates new metrics tailored to the specific characteristics of nuclei instance segmentation, offering a more comprehensive assessment of an algorithm's capabilities in detection, segmentation accuracy, and topological structure preservation. By rigorously testing and ranking existing mainstream algorithms on the fused dataset, NucFuseRank aims to provide a trustworthy and reproducible benchmark, assisting researchers in identifying state-of-the-art methods, understanding their strengths and weaknesses, and guiding future algorithm development.