PhD candidate 'Al-powered handheld ultrasound for prenatal diagnosis'

Every day more than 800 women die as a consequence of their pregnancy, of which the vast majority occur in low-income countries. Ultrasound imaging can be used to detect maternal risk factors and is recommended by the World Health Organization. Unfortunately, ultrasound is barely used in these countries. This is caused by the lack of trained sonographers that can perform prenatal scanning and the high cost of ultrasound equipment.

The latter obstacle is alleviated by handheld ultrasound devices that are now becoming available on the market for relatively low prices. Most of these devices can be connected to a smartphone or tablet. To obviate the need of a trained sonographer, we propose to combine a standardize acquisition protocol (consisting of several sweeps over the belly of the pregnant woman) with real-time feedback by deep learning algorithms followed by automated interpretation with deep learning. In the past, algorithms were developed in our group for automated to detect twin pregnancies, estimate gestational age, determine fetal presentation and determine placenta location. All developed algorithms were ported to an Android based app called the BabyChecker and our industry partner Delft Imaging has started data collection and field tests in several African countries. Delft Imaging is funding this Ph.D. position.

Tasks and responsibilities
You will work on extending the capabilities of the BabyChecker with:

• Improving the real-time feedback to guide during the acquisition of the sweeps, so that acquisition errors can be prevented or immediately corrected.
• Developing and validating several new automated algorithms for example for the detection of fetal viability, detection of placenta previa, and estimation of the amount of amniotic fluid.
• Improve the performance of the existing algorithms.
• Increase the robustness of the deep learning algorithms so they provide accurate results with data from multiple types of hand-held ultrasound devices that are available on the market today.

All algorithms should be computationally efficient so they can be run on a smartphone. We expect you'll be using and extending state-of-the-art frameworks for deep learning on low-cost devices. We also expect you to be active in data collection and field testing and interested to travel to Africa and other sites where the BabyChecker is used.