Background

During pregnancy, ultrasound imaging is used to measure fetal biometrics. One of these measurements is the fetal head circumference (HC). The HC can be used to estimate the gestational age and monitor growth of the fetus. The HC is measured in a specific cross section of the fetal head, which is called the standard plane. The dataset for this challenge contains a total of 1334 two-dimensional (2D) ultrasound images of the standard plane that can be used to measure the HC. This challenge makes it possible to compare developed algorithms for automated measurement of fetal head circumference in 2D ultrasound images. 

Detailed description of the data

The data for this challenge can be downloaded from Zenodo, DOI 10.5281/zenodo.1322001 

The data is divided into a training set of 999 images and a test set of 335 images. The size of each 2D ultrasound image is 800 by 540 pixels with a pixel size ranging from 0.052 to 0.326 mm. The pixel size for each image can be found in the csv files: ‘training_set_pixel_size_and_HC.csv’ and ‘test_set_pixel_size.csv’. The training set also includes an image with the manual annotation of the head circumference for each HC, which was made by a trained sonographer. The csv file 'training_set_pixel_size_and_HC.csv ' includes the head circumference measurement (in millimeters) for each annotated HC in the training set. All filenames start with a number. There are 999 images in the trainingset, but the filenames only go to 805. Some ultrasound images were made during the same echoscopic examination and have therefore a very similar appearance. These images have an additional number in the filename in between "_" and "HC" (for example 010_HC.png and 010_2HC.png).

Challenge

This challenge is aimed to design an algorithm that can automatically measure the fetal head circumference given a 2D ultrasound image. You can use the 999 ultrasound images and the corresponding annotations in the training set to develop an algorithm that can automatically measure the HC. The developed algorithm should be applied to the independent test set of 335 images. You can upload your results on the test set under the Submit tab. The results will be immediately evaluated upon submission and displayed on the leaderboard under the Results tab. This gives you the possibility to compare your algorithm to other algorithms on the same independent test set.

Cite as

When referencing this challange, please include a bibliographical reference to the paper and the dataset below:
- Thomas L. A. van den Heuvel, Dagmar de Bruijn, Chris L. de Korte and Bram van Ginneken. Automated measurement of fetal head circumference using 2D ultrasound images. PloS one, 13.8 (2018): e0200412.
- Thomas L. A. van den Heuvel, Dagmar de Bruijn, Chris L. de Korte and Bram van Ginneken. Automated measurement of fetal head circumference using 2D ultrasound images [Data set]. Zenodo. http://doi.org/10.5281/zenodo.1322001

Submission instructions

The results should be submitted as a csv file which contains 6 columns and 336 rows. The first row should be:

  filename,center_x_mm,center_y_mm,semi_axes_a_mm,semi_axes_b_mm,angle_rad

Each of the other rows should describe the ellipse of one of the 335 images in the test set. The filename should include ".png" (for example: 001_HC.png). There are five values that describe the ellipse. Be aware that the angle should be given in radians and the other four values in millimeters. Below an example image (388_HC.png from the training set with the expert annotation) the csv file for this ellipse would state:

388_HC.png 72.378073183 60.803647248 32.482252220 23.374422688 0.061252332740