Pathological full-slice image segmentation algorithm based on cascading idea

The invention discloses a pathological full-slice image segmentation algorithm based on a cascading idea, and the algorithm comprises two U-Net structures. A first network is trained by employing samples collected at a low resolution, an area which is easy to segment is filtered, and an approximate cancer area segmentation result is obtained; and a second network optimizes the cancer area segmentation result obtained by the first network. According to the method, the digital pathological image segmentation precision is improved, and the test time is shortened. 本发明公开了一种基于级联思想的病理全切片图像分割算法，包括：两个U-Net结构，第一个网络用低分辨率下采集的样本进行训练，滤除易于分割的区域，得到大致的癌症区域分割结果；第二个网络优化第一个网络得到的癌症区域分割结果。本发明不仅提升了数字病理图像分割精度，而且缩短了测试时间。