OMG-Octo Atypical: A refinement of the original OMG-Octo database to incorporate atypical mitoses

In this study, we pre-trained our model on the MIDOG 2025 atypical mitoses database (AMi-Br: 832 atypical and 2,888 normal mitotic figures and MIDOG-Atypical (1748 atypical and 10191 normal mitoses), From this model, we inferred to find atypical mitoses within our OMG-Octo dataset and found 10678 potential candidates. Dedicated pathologists reviewed 3024 of them (1378 atypical, 394 apoptotic, 379 normal MF, 399 Noise and 462 Unsure). The results are uploaded here under a CC BY.0 NC ND 4.0 International License. Publications and upates will follow.本研究中，我们基于MIDOG 2025非典型有丝分裂数据库开展模型预训练，该数据库包含AMi-Br数据集（832个非典型有丝分裂样本、2888个正常有丝分裂图像）与MIDOG-Atypical数据集（1748个非典型有丝分裂样本、10191个正常有丝分裂样本）。随后，我们将预训练后的模型应用于OMG-Octo数据集以识别非典型有丝分裂，共得到10678个潜在候选样本。由专业病理学家对其中3024个候选样本进行标注复核，复核结果如下：1378个非典型有丝分裂样本、394个凋亡相关样本、379个正常有丝分裂图像、399个噪声样本，以及462个难以判定的样本。本数据集成果已基于CC BY-NC-ND 4.0国际许可协议上传至此处。后续将发布相关研究成果与数据集更新内容。