Obstetrics Gynaecology



Each day, more than 800 women die as a consequence of their pregnancy. 99% of these maternal deaths occur in resource-limited settings. Ultrasound imaging is commonly used to detect maternal risk factors and even though the world health organization has recommended the usage of ultrasound imaging, this technique is barely used in resource-limited settings due to the severe shortage of well-trained medical personnel.

Current attempts to introduce prenatal ultrasound in resource-limited settings has focused on training sonographers to both acquire and interpret prenatal ultrasound. This training takes several months up to two years, which is a time-consuming task that is hampering wide-spread of prenatal ultrasound screening.

Automated low-cost ultrasound: improving antenatal care in resource-limited settings.


To make prenatal ultrasound screening available in resource-limited settings we combine a low-cost smartphone based ultrasound device with deep learning algorithms that assist the user in obtaining a standardized acquisition protocol. This protocol can be taught to any healthcare worker within two hours of training.


The deep learning algorithms interpret the acquired ultrasound images which automatically detects twin pregnancies, estimates gestational age and determines fetal presentation. The algorithms run locally on the smartphone and are able to process the acquired ultrasound images in real-time.

Us frame classification.jpg
Example of the B-mode images of a standardized ultrasound acquisition protocol for prenatal screening with the automated frame classification below.)



The female pelvic floor (PF) muscles provide support to the pelvic organs. During delivery, some of these muscles have to stretch up to three times their original length to allow for the passage of the baby, leading frequently to damage and consequently later life PF dysfunctions. Three-dimensional (3D) ultrasound (US) imaging can be used to image these muscles and to diagnose these damages by assessing quantitative, geometrical and functional information of the muscles through strain imaging. In our work we have developed 3D US strain imaging of the PF muscles and explored its application on the puborectalis muscle (PRM), which is one of the major PF muscles. Our work is funded by the GYNIUS grant from NWO.


In our work, US data from female pelvic floor is obtained by the radiologist through transperineal US. Data acquisition takes place while the muscle voluntarily deformed, for example, from rest to contraction. The 3D US volumes are then used to measure the displacement and deformation of the muscle. For calculating the displacement estimations in 3D, the algorithm developed in our group is used. Details of this algorithm and strain measurement can be found here [1]. Tracking of the moving muscle is performed to measure the displacements between two adjacent US volumes.


The displacement/dislocation and deformation in the muscle PRM can be shown through the below three images. In the first figure (Fig. 1), the muscle (blue) is at rest and is shown in 3D as well as the sagittal, axial and coronal anatomical views. In the second figure (Fig. 2), the muscle (red) is shown when it is deformed at contraction, also in 3D and the three anatomical views. The third figure (Fig. 3), shows the muscle at rest and contraction superimposed on each other, to emphasize the deformation seen in the muscle, revealed through tracking.

GYNIUS Tracking.jpg
Figure 1-3: PRM shown in 3D US grid and anatomical views for rest (blue) and contraction (red)

Strain imaging of the PRM is performed to have a better understanding of the functional information of the muscle. This functional information is obtained by measuring principal strain in the muscle. From the results of principal strain a non-damaged muscle can be distinguished from a damaged one. For example, a muscle which has no damage in it shows almost uniformly distributed strain throughout the contracted muscle (Fig. 4). Whereas when a certain damaged muscle shows contraction (Fig. 5), the strain is very high in one end of the muscle (shown as the dark red area). But the other end of the muscle shows very less strain. This means that the healthy end of the muscle is pulling the damaged end of the muscle which cannot move by itself.

GYNIUS Results.jpg
Figure 4-5: Percentage of strain in contracted PRMs (with and without damage)


  • GYNIUS: Gynaecological Imaging using 3D Ultrasound (Dutch Technology Foundation STW and Philips, active)Project Shreya?
  • BabyViewer (Eurostars project, finished)
  • Finding Obstetric Complications using UltraSound and Artificial Intelligence (NWO-TTW Demonstrator, active)

Our people

PhD candidates

Scientific staff

Visiting researchers

Key publications

  • T.L.A. van den Heuvel, H. Petros, S. Santini, C.L. de Korte and B. van Ginneken. "Automated Fetal Head Detection and Circumference Estimation from Free-Hand Ultrasound Sweeps Using Deep Learning in Resource-Limited Countries", Ultrasound in Medicine and Biology 2019;45(3):773-785. Abstract/PDF DOI PMID

  • T.L.A. van den Heuvel, D. de Bruijn, C.L. de Korte and B. van Ginneken. "Automated measurement of fetal head circumference using 2D ultrasound images", PLoS One 2018;13(8). Abstract/PDF DOI PMID

  • T.L.A. van den Heuvel, D. de Bruijn, D. Moens-van de Moesdijk, A. Beverdam, B. van Ginneken and C.L. de Korte. "Comparison Study of Low-Cost Ultrasound Devices for Estimation of Gestational Age in Resource-Limited Countries", Ultrasound in Medicine and Biology 2018;44(11):2250-2260. Abstract/PDF DOI PMID

  • T.L.A. van den Heuvel, D.J. Graham, K.J. Smith, C.L. de Korte and J.A. Neasham. "Development of a Low-Cost Medical Ultrasound Scanner Using a Monostatic Synthetic Aperture", IEEE Transactions on Biomedical Circuits and Systems 2017;11(4):849-857. Abstract/PDF DOI PMID