标题： 在之前基础模型未解决的具有挑战性的肾脏病理病例上评估新的AI细胞基础模型 显示英文标题

标题： Evaluating New AI Cell Foundation Models on Challenging Kidney Pathology Cases Unaddressed by Previous Foundation Models

摘要： 准确的细胞核分割对于肾脏病理学中的下游任务至关重要，但由于肾组织的形态多样性和成像变异性，这仍然是一个主要挑战。 尽管我们之前的工作已经评估了该领域早期生成的AI细胞基础模型，但最近的细胞基础模型的有效性仍不明确。 在本研究中，我们在一个人机交互评分框架内，使用一组多样化的大规模肾脏图像块，对先进的AI细胞基础模型（2025年）进行了基准测试，包括CellViT++变体和Cellpose-SAM，与三个在2024年之前开发的广泛使用的细胞基础模型进行比较。 我们进一步进行了基于融合的集成评估和模型一致性分析，以评估不同模型的分割能力。 我们的结果表明，CellViT++ [Virchow] 在经过筛选的2,091个具有挑战性的样本集上，有40.3%的预测被评为“良好”，其表现优于所有先前模型。 此外，我们的融合模型实现了62.2%的“良好”预测，并且只有0.4%的“差”，显著减少了分割错误。 值得注意的是，融合模型（2025年）成功解决了我们之前研究中未解决的大部分具有挑战性的情况。 这些发现展示了AI细胞基础模型在肾脏病理学中的潜力，并提供了一个经过筛选的具有挑战性的样本数据集，以支持未来针对肾脏的模型优化。 显示英文摘要

摘要： Accurate cell nuclei segmentation is critical for downstream tasks in kidney pathology and remains a major challenge due to the morphological diversity and imaging variability of renal tissues. While our prior work has evaluated early-generation AI cell foundation models in this domain, the effectiveness of recent cell foundation models remains unclear. In this study, we benchmark advanced AI cell foundation models (2025), including CellViT++ variants and Cellpose-SAM, against three widely used cell foundation models developed prior to 2024, using a diverse large-scale set of kidney image patches within a human-in-the-loop rating framework. We further performed fusion-based ensemble evaluation and model agreement analysis to assess the segmentation capabilities of the different models. Our results show that CellViT++ [Virchow] yields the highest standalone performance with 40.3% of predictions rated as "Good" on a curated set of 2,091 challenging samples, outperforming all prior models. In addition, our fused model achieves 62.2% "Good" predictions and only 0.4% "Bad", substantially reducing segmentation errors. Notably, the fusion model (2025) successfully resolved the majority of challenging cases that remained unaddressed in our previous study. These findings demonstrate the potential of AI cell foundation model development in renal pathology and provide a curated dataset of challenging samples to support future kidney-specific model refinement.