Scaffold-A549: A Benchmark 3D Fluorescence Image Dataset for Unsupervised Nuclei Segmentation

A general trend of nuclei segmentation is the transition from two-dimensional to three-dimensional nuclei segmentation and from traditional image processing methods to data-driven cognitively inspired methods. Existing nuclei segmentation datasets do not meet this trend: They either do not contain enough samples for training the deep learning model or not contain challenging 3D structure. Thus, large-scale datasets are critically demanded for nuclei segmentation tasks. In this paper, we introduce a new benchmark nuclei segmentation dataset termed as Scaffold-A549 for 3D cell culture on bio-scaffold. The A549 human non-small cell lung cancer cells are seeded in the bio-scaffold for cell culture and the samples with different density of nuclei are captured using confocal laser scanning microscope at the first, third, and eighth culture day. A total of 21 3D images are collected containing more than 10,000 nucleus and each of the images containing more than 800 nucleus are annotated manually for evaluation. Scaffold-A549 presents one large, diverse, challenging, and publicly available dataset and can be widely used for the research on 3D unsupervised nuclei segmentation.

Duke Scholars

Author Kaizhu Huang DKU Faculty