datatorch

/

Annotated metaphase chromosome images (Tseng et al. 2023)

Public

Updated a day ago

microscopycytogenetics
Label your own images with these labels

Files

About

Annotated metaphase chromosome images (Tseng et al. 2023)

Clinical G-banded metaphase cell images from the prenatal cytogenetics laboratory of Taichung Veterans General Hospital, published with chromosome bounding boxes by Tseng et al. in Scientific Data and distributed through the Cell Image Library as CIL:54816 (doi:10.7295/W9CIL54816). The paper states the images were collected with review-board approval and patient consent. The original Pascal VOC boxes are imported here as browsable bounding-box annotations in two sets: one with every chromosome labelled by its karyotype class, one with every chromosome under a single class.

Contents

Images6,993 JPEGs: 4,997 in images/boxes24/ (24-class set) and 1,996 in images/single/ (single-class set)
Annotations321,838 bounding boxes: 229,714 in the 24-class set and 92,124 in the single-class set (43 to 57 per image, median 46)
Classes24-class set: A1 to G22 (autosomes 1 to 22 by Denver group), X, Y. Single-class set: chromosomes
Datasetmetaphase-chromosomes-cil54816-v1
SplitSource train/test lists, re-keyed: tables/train.txt (3,742 rows) and tables/test.txt (1,249 rows), one boxes24_id,single_id pair per row
ImagingBright-field G-banded metaphase spreads, grayscale-looking JPEG, 360 to 1,360 px wide and 383 to 1,024 px high (median about 706 x 687)
Extra filestables/normal.csv, tables/number_abnormalities.csv, tables/structural_abnormalities.csv (karyotype per image), tables/diff_image.txt (images the authors rated difficult), annotations/pascal_voc_xml.tar.gz (the original-style VOC XML)

1,843 images appear in both sets as byte-identical files under two different ids. The tables pair 1,833 of them through their boxes24_file/single_file columns.

Using it

Browse any image to see the boxes. To download everything (free account required):

  • ZIP: use the Download button on this project.
  • Python SDK: pip install datatorch, then:
datatorch login
datatorch pull datatorch/metaphase-chromosomes-cil54816

Annotations export in COCO format through the project's export schema.

Conversion notes

  • Labels. The source ships one Pascal VOC XML per image (segmented is 0 in every file, so these are boxes, not masks). Boxes were converted from xmin, ymin, xmax, ymax to COCO x, y, w, h and imported as bounding boxes with no segmentation. The coordinate order was checked against image sizes and by drawing the boxes over the chromosomes. 77 boxes in the 24-class set and 118 in the single-class set extended past the image edge and were clipped to it.
  • Renamed files. Every image was renamed to a randomly shuffled id (boxes24_NNNNNN.jpg, single_NNNNNN.jpg). The source folder 24_chromosomes_object is images/boxes24/ here and single_chromosomes_object is images/single/. The mapping from source names is not published.
  • Annotation files. The path field was removed from every XML and filename was set to the new id. The folder field (chromosome_original) is kept.
  • Tables re-keyed. The three CSVs and train.txt, test.txt and diff_image.txt were rewritten to the new ids. The first column (source image name) became two columns, boxes24_file and single_file, because one source name can refer to an image in each set.
  • Image metadata removed. EXIF/XMP (APP1), other APPn and comment segments were dropped from every JPEG. The pixel data is byte-for-byte unchanged, and decoded pixels were checked to be identical to the source.
  • Five images left out. Five source images (seven files, counting the copies in both sets) were not included because a hospital case code was readable in the image. As a result the 24-class set has 4,997 of 5,000 images and the single-class set has 1,996 of 2,000. Rows naming these images were dropped from the tables: 3 from normal.csv, 2 from train.txt and 1 from test.txt.
  • Lab labels in pixels. The other images were not searched or edited. A small share of them may carry a lab label in the pixels, as the source published them.
  • Weights not included. The two YOLOv4 weight files in the source (weight/*.pt) are not part of this project.
  • Source inconsistencies, kept as found. Rows naming an image that is not in the archive were dropped: 2 from normal.csv, 4 from number_abnormalities.csv, 2 from structural_abnormalities.csv and 6 from train.txt. Six images in the 24-class set are in neither split list. The source spells its folders JEPG and anntations. The Cell Image Library record lists the cell type as "epithelial cell" and the year as 2022. The paper was published in 2023.
  • One image (images/boxes24/boxes24_000001.jpg) is recorded with the file type text/plain on the server although it is a JPEG and its boxes imported normally.

License and citation

CC BY 3.0 (Creative Commons Attribution 3.0), as stated on the Cell Image Library record. You may share and adapt the data with attribution, including the changes listed above. If you use this data, please cite:

Tseng, J.J., Lu, C.H., Li, J.Z., Lai, H.Y., Chen, M.H., Cheng, F.Y., Kuo, C.E. An Open Dataset of Annotated Metaphase Cell Images for Chromosome Identification. Scientific Data 10, 104 (2023). doi:10.1038/s41597-023-02003-7

and the dataset:

Lu, C.-H., Kuo, C.-E. & Tseng, J.-J. CIL:54816. Cell Image Library (2022). doi:10.7295/W9CIL54816

Source: CIL:54816, The Cell Image Library.

Related

More public datasets

All in microscopy
  • Nuclei Segmentation (DSB2018)datatorch/nuclei-segmentation-dsb2018
  • MCF7 and A549 NF-kB translocation (BBBC014)datatorch/translocation-bbbc014
  • iFISH chromosome spotting (IDR idr0123)datatorch/ifish-chromosome-spotting-idr0123
  • Deep-Fish-Bone zebrafish larval skeleton segmentationdatatorch/deep-fish-bone-zebrafish
  • AxonPath Myelinated Nerve Fibre Segmentationdatatorch/axonpath-myelinated-fibres

Description

Annotated metaphase chromosome images from Tseng et al. 2023 (Cell Image Library CIL:54816): clinical prenatal G-banded metaphase cell images from the cytogenetics laboratory of Taichung Veterans General Hospital, collected with review-board approval and patient consent as the paper states. 6,993 images with bounding boxes (boxes, not masks) in two sets: 4,997 images with 229,714 boxes in 24 chromosome classes (A1-G22, X, Y) and 1,996 images with 92,124 boxes in a single class (chromosomes); 1,843 images appear in both sets. Changes from the source: files renamed to sequential ids, the path field removed from the annotation files, label tables and split lists re-keyed to the new names, image metadata removed, model weight files not included, and five images left out because a lab label was readable in the image. A small share of the remaining images may carry a lab label in the pixels, as published by the source. License: CC BY 3.0. Source: http://www.cellimagelibrary.org/images/54816. Please cite: Tseng JJ, Lu CH, Li JZ, Lai HY, Chen MH, Cheng FY, Kuo CE. An Open Dataset of Annotated Metaphase Cell Images for Chromosome Identification. Sci Data 10, 104 (2023). doi:10.1038/s41597-023-02003-7. Dataset: CIL:54816, doi:10.7295/W9CIL54816.

Members

0
View all members →

Progress

100% labeled

6,993 of 7,007 files

Size

525.1 MB

Datasets

1

Labels

25

Annotations

321,838