Journal: GigaScience

Article Title: CellBinDB: a large-scale multimodal annotated dataset for cell segmentation with benchmarking of universal models

doi: 10.1093/gigascience/giaf069

Figure Lengend Snippet: CellBinDB overview. (A) Distribution of staining types in CellBinDB. (B) Distribution of tissue types in CellBinDB, where tissue types with fewer than 10 samples are included in "other"; for details, see . (C) Examples of CellBinDB images with scale bar and instance ground-truth annotations, from left to right: column 1, ssDNA; column 2, DAPI; column 3, H&E; column 4, mIF; column 5, 10x Genomics DAPI; column 6, 10x Genomics H&E. The first row provides the original microscope images, and the second contains the instance annotation masks. (D) Scatterplot of t-SNE demonstrates the diverse spread of data by different staining types and sources. (E) Scatterplot of t-SNE demonstrates the diversity of CellBinDB compared to previous datasets. (F) The number of manual and semiautomatic annotations in CellBinDB. (G) The dataset annotation process includes 4 steps: 1, model annotation; 2, annotation team modification/reannotation (depends on the model annotation results); 3, expert review, go to the next step if the annotations are correct, otherwise return to the second step for modification; and 4, add the original image and the 2 masks to the dataset.

Article Snippet: Whole-slide images (WSIs) were generated by (i) a STOmics Microscope Go Optical equipped with Scanner Version 1.2.2, using 10×/0.75 NA and 20×/0.5 NA objective and Go Optical Scanner Ximea Mc124 for grayscale images and Go Optical Scanner Ximea Mc050 for RGB images, and (ii) a Motic PA53 FS6 microscope equipped with PA53Scanner 1.0.0.14, using 10×/0.75 NA objective and PA53 FS6 SCAN S5LITE MONO for grayscale images and PA53 FS6 SCAN S5LITE for RGB images.

Techniques: Staining, Microscopy, Modification