Sperm chromatin dispersion bright field images
Description
This dataset is constructed with 130 images and annotations for bright field microscopic images (20x objective) from results of sperm chromatin dispersion test. Sperm chromatin dispersion test (SCD) is a microscopic method for determining sperm DNA fragmentation, sperm DNA is dispersed into the gel on the slide by the lysing buffer. Sperms with higher DNA fragmentation produces more single-stranded DNA in the nucleus, and preventing the dispersion process. Wright-Giemsa staining is performed to visualize the DNA on the slide, thus sperms with DNA damage is revealed by the relatively smaller halos. For steps to utilize this dataset, see the descriptions below.
Files
Steps to reproduce
The dataset is constructed with two parts: (1) the bright-field images with numeral file names (2) the corresponding annotations in the form of tuples separated by semicolons in the text files. For example, the image file /Images/202.jpg is linked with /Annotations/202.txt, and the format of the annotation text file may be: ``` 331.00;241.00;278.00;241.00;30.00;46.99 998.00;592.00;945.00;566.00;48.38;52.46 ... ``` Each line is composed of 6 numerals that marks the geometric locations of each sperm in the picture. [0] Tail X [1] Tail Y [2] Head X [3] Head Y (in pixels, to the corresponding images sized 3072 * 2048 [4] Length of the minor axis (2 * semi-minor axis) [5] The maximal thickness of the halo in the direction of the minor axis The images are sized 3072 * 2048, however, the network can only accept 512 * 512 images as input, thus, the dataset must be preprocessed to the target size. In our research, you can execute a preprocessor utility to generate the dataset in '<data-set-name>.d' file (actually a zip file, containing clipped 512 * 512 images and its notations in numpy format) ``` python preprocess.py -i "<path_to_dataset>/Images/" -a "<path_to_dataset>/Annotations/" -s 512 -t 0.5 -m "0 0 0 0" -p "datasets.scds.scdx16p100" "scdx16p100.d" ``` The source is published on Github: https://github.com/Xornent/scd-resnet/, largely based on the implementation of CenterNet.