Normal Colon Crypt Dataset for U-shaped Crypt Segmentation

Dataset Overview: In the literature, some colon tissue H&E and IHC datasets exist, but none of them focus on normal colon crypts. This is a first-of-its-kind curated and annotated dataset focused on normal U-shaped crypts. The dataset consists of extracted regions of interest (ROIs) from whole-slide images, along with color-normalized versions and annotated masks. The dataset is used to build a gland instance segmentation model. Dataset Content: Colon_Crypt_Segmentation_Dataset/ │── training/ annotated images: crypt and gland masks │ │── raw_patches_wsi/ images in their original colours │ │ │── train/ valid/ test/ │ │ │ │──Images/ one PNG per image │ │ │ │──Annotation/ crypt masks: one PNG per crypt, named <image name>_<n>.png │ │ │ └──Annotation_gland/ gland masks: one PNG per gland, named <image name>_<n>.png │ └── color_normalized/ colour-normalized images, same layout as raw_patches_wsi/ │ └── color_reference.png reference image used for the colour normalization└── Inference/ images without annotations used for segmentation model inference │──raw_patches_wsi/ <staining_protocol>/<sample>/50HD####.JPG (original camera photos) └── color_normalized/ <staining_protocol>/<sample>/50HD####.png (color-normalized and hough transformed) Image Count Crypt masks Gland masks training/raw_patches_wsi 293 1044 1445 training/color_normalized 291 1044 1445 Inference/raw_patches_wsi 648 unlabeled unlabeled Inference/color_normalized 645 unlabeled unlabeled The whole-slide scanning technique is used to acquire high-resolution images from human and mouse colon tissue slides. For each slide, the standard protocol is followed for the gene targets. Specifically, the slides are stained with CDX2, CAI, MUC2, KRT20, and SLC26A3. The images are saved in different formats, such as NDPI, TIFF, SVS, and CZI. The colon glands appear as U- or O-shaped in each slide, depending on whether they are cut horizontally or vertically. Vertical cross-sections have been applied to each slide to obtain more U-shaped data. Normal crypt regions have been extracted using QuPath software under the supervision of biologists. In total, 1,044 U-shaped crypts and 1,445 O-shaped crypts were annotated using GIMP software from 291 slide images. These images were split into training, testing, and validation sets following the general rule. Microscope Details: Nikon A1R Confocal TIRF STORM Microscope Nikon Ti microscope with integrated autofocus (Perfect Focus) that maintains focus when using glass slides, plastic culture dishes, or multiwell plates. Automated XY and Z stage (including a fast piezo Z). Andor iXon+DU897, EMCCD (512x512 pixels with readout of 30 full frames per second) for low-light applications such as TIRF, FRAP, and single-molecule imaging applications such as STORM. Andor high-speed sCMOS camera (Zyla 5.5 megapixel with readout of 100 full frames per second) for applications requiring a higher resolution and speed. A1R hybrid confocal scanner incorporates two independent galvo scanning systems, a high-resolution (4096x4096) scanner, and a high-speed resonant scanner (512x512) for rapid confocal scanning of live cells. LU4 four-laser AOTF unit with 405, 488, 561, and 647 lasers. Lumencore solid-state light source for wide-field illumination and fast switching. Plan Apo 10, 20 and 40X dry objectives, and 60 and 100X high NA (1.49NA) TIRF objectives. Provided with heating and carbon dioxide incubation for live-cell work. Olympus VS200 Slide Scanner - Fluorescence and Brightfield Generating brightfield and multi-color fluorescent images of entire slides or tissue sections automatically 7 fluorescent colors Fixed cells or sections on a standard 1 x 3 slide with a coverslip or 2 x 3 slides Can scan up to 210 slides automatically Fluorescent light from an LED lamp filtered for emission or brightfield light produces an image by tiling specified areas of a slide to generate montage reconstructions. 7 fluorescent color channels and brightfield: DAPI - Ex 345 Em 432/36 FITC - Ex 494 Em 515/30 TRITC - Ex 555 Em 600/31 CY5 - Ex 625 Em 685/42 Cy7 - Ex 743 Em 810/81 CFP - Ex 458 Em 482/25 YFP - Ex 513 Em 544/25 Motorized X-Y capability for slide scanning Microscope objective: 2x, 4x, 20x (.80 NA), 40x Air Applications: Segmentation model training and fine-tuning Digital Pathology Deep learning-based image analysis