MP: Real-Time Accurate Measurement of Placental Volume by Ultrasound

In pregnancy care, it is useful to identify pregnancies that have an increased risk of complications as early as possible. In this way, these women can be followed more closely, and preventive measures can be taken if necessary. Today, at least one ultrasound examination of the fetus is typically performed in the course of the pregnancy. Here, several features of the pregnancy may be evaluated, such as the number of fetuses, the location of the placenta and the fetus' development and anatomy. Over the last 10-15 years, it has also been established an association between the volume of the placenta and the risk of complications. However, in order to measure this volume during the pregnancy, one must have good three-dimensional (3D) images. Ultrasound probes for 3D imaging are becoming more common, but these cover a limited area, and it is often not possible to fit the entire placenta within one image. In this project, we have investigated a method for measuring the placental volume automatically by stitching together two-dimensional (2D) images from a normal ultrasound probe. This was done by equipping an ultrasound probe with a position sensor, and then imaging a large number of placentas while measuring the position of each individual ultrasound image. Using modern machine learning methods we can then teach a computer program how 2D images can be combined into a 3D image based solely on what is seen in the images. These 3D images can then be used to calculate the volume of the placenta. So far, we have collected both ultrasound and MR images from 43 pregnant women and used these to verify the volume measurements and develop a machine learning method for automatic identification of the placenta in ultrasound images. These data and methods will be further integrated and developed in a new project. In the long term, the method we are developing can be used on any ultrasound scanner, and it can therefore help identify many high-risk pregnancies that go unnoticed today. This is necessary to be able to take measures that could potentially prevent damage to the fetus or, in the worst case, fetal death. The method can thus improve maternal care all over the world.

The results from this project indicate that it could be possible both to measure the placental volume accurately using ordinary 2D ultrasound probes, and to automate the measurements using machine learning methods. In the developed regions of the world, there is an estimated 13 million births per year, from which Norway accounts for almost 60 000. In these regions, ultrasound examinations are established as a standard part of maternity care offered to all pregnant women. With the proposed method, reliable measurements of placental volume can potentially be incorporated in these examinations. This would enable the identification of many high-risk pregnancies that today go unnoticed, and thus reduce the occurrence of adverse pregnancy outcomes. All maternity care units in these countries, and also in many countries in the less developed regions of the world, are therefore potential users of this technology.