人工智能培训

搜索

机器学习论文:CIA-NET:具有轮廓感知信息聚合的强大核实例分段(CIA-Net: Robust Nuclei Instance Segmentation w

[复制链接]
einter 发表于 2019-3-15 12:37:45 | 显示全部楼层 |阅读模式
einter 2019-3-15 12:37:45 542 0 显示全部楼层
机器学习论文:CIA-NET:具有轮廓感知信息聚合的强大核实例分段(CIA-Net: Robust Nuclei Instance Segmentation with Contour-aware  Information Aggregation)准确分割细胞核实例是计算机辅助图像分析中的关键步骤,以提取用于细胞估计和随访诊断以及治疗的丰富特征。虽然它仍然具有挑战性,因为核簇的广泛存在,以及不同器官的大的形态变化使得核实例分割易于过度/不足分割。此外,不可避免的主观注释和错误标记阻止了网络从可靠样本中学习,并最终降低了强大地分割非中心核的泛化能力。为了解决这些问题,我们提出了一种新颖的深度神经网络,即轮廓感知信息聚合网络(CIA-Net),在两个任务专用解码器之间具有多级信息聚合模块。它不是独立的解码器,而是通过双向聚合特定于任务的特征来利用核和轮廓之间的空间和纹理依赖性的优点。此外,我们提出了一种新的平滑截断的losst,它调制损耗以减少异常值的扰动。因此,网络可以专注于从可靠和信息丰富的样本中学习,这样可以提高泛化能力。关于2018MICCAI多器官核分割挑战的实验验证了我们提出的方法的有效性,超过了所有其他35个竞争团队的显着差距。
Accurate segmenting nuclei instances is a crucial step in computer-aidedimage analysis to extract rich features for cellular estimation and followingdiagnosis as well as treatment.While it still remains challenging because thewide existence of nuclei clusters, along with the large morphological variancesamong different organs make nuclei instance segmentation susceptible toover-/under-segmentation.Additionally, the inevitably subjective annotatingand mislabeling prevent the network learning from reliable samples andeventually reduce the generalization capability for robustly segmenting unseenorgan nuclei.To address these issues, we propose a novel deep neural network,namely Contour-aware Informative Aggregation Network (CIA-Net) with multi-levelinformation aggregation module between two task-specific decoders.Rather thanindependent decoders, it leverages the merit of spatial and texturedependencies between nuclei and contour by bi-directionally aggregatingtask-specific features.Furthermore, we proposed a novel smooth truncated lossthat modulates losses to reduce the perturbation from outliers.Consequently,the network can focus on learning from reliable and informative samples, whichinherently improves the generalization capability.Experiments on the 2018MICCAI challenge of Multi-Organ-Nuclei-Segmentation validated the effectivenessof our proposed method, surpassing all the other 35 competitive teams by asignificant margin.机器学习论文:CIA-NET:具有轮廓感知信息聚合的强大核实例分段(CIA-Net: Robust Nuclei Instance Segmentation with Contour-aware  Information Aggregation) R1hdwZ5xcUHvDr6R.jpg
URL地址:https://arxiv.org/abs/1903.05358     ----pdf下载地址:https://arxiv.org/pdf/1903.05358    ----机器学习论文:CIA-NET:具有轮廓感知信息聚合的强大核实例分段(CIA-Net: Robust Nuclei Instance Segmentation with Contour-aware  Information Aggregation)
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则 返回列表 发新帖

einter当前离线
新手上路

查看:542 | 回复:0

快速回复 返回顶部 返回列表