Data for: Fast and robust segmentation of cell images by self-supervised learning

Two datasets consisting of white blood cell (WBC) images were tested in our experiments. These two datasets are chosen for our experiments, as they are significantly different from each other in terms of the image color, cell shape, background, etc., which can better evaluate the robustness of the proposed approach. The ground truth segmentation results are manually sketched by domain experts, where the nuclei, cytoplasms and background including red blood cells are marked in white, gray and black respectively. We also submitted the segmentation results by our approach, where the whole WBC region are marked in white and the others are marked in black. Dataset 1 was obtained from Jiangxi Tecom Science Corporation, China. It contains three hundred 120×120 images of WBCs and their color depth is 24 bits. The images of Dataset 1 were taken by a Motic Moticam Pro 252A optical microscope camera with a N800-D motorized auto-focus microscope, and the blood smears were processed with a newly-developed hematology reagent for rapid WBC staining. The overall background of images in Dataset one looks yellow. Dataset 2 consists of one hundred 300×300 color images, which were collected from the CellaVision blog (http://blog.cellavision.com). The cell images are generally purple and may contain many red blood cells around the white blood cells, which is quite different from Dataset 1.

本实验共测试了两组白细胞（white blood cell, WBC）图像数据集。选用这两组数据集开展实验，是因为二者在图像色彩、细胞形态、背景等维度均存在显著差异，可更充分地评估所提方法的鲁棒性。分割真值（ground truth）均由领域专家手动绘制，其中细胞核、细胞质以及包含红细胞在内的背景分别以白色、灰色和黑色标记。本研究同时提交了所提方法生成的分割结果：整个白细胞区域以白色标记，其余区域以黑色标记。数据集1采集自中国江西泰豪科技股份有限公司（Jiangxi Tecom Science Corporation），包含300张分辨率为120×120的白细胞图像，图像色彩深度为24比特。该数据集的图像由搭载N800-D电动自动对焦显微镜的Motic Moticam Pro 252A光学显微相机拍摄，血涂片采用新型快速白细胞染色血液学试剂处理。数据集1的整体背景呈黄色调。数据集2包含100张分辨率为300×300的彩色图像，采集自CellaVision博客（http://blog.cellavision.com）。该数据集内的细胞图像整体偏紫色，且白细胞周围常存在大量红细胞，与数据集1差异显著。