Intelligent Imaging

for Prenatal Diagnosis

Imaging Improves Fetal Brain Diagnosis

Impaired fetal circulation, including congenital heart disease (CHD) and intrauterine fetal growth restriction (IUGR), may lead to abnormal development of the fetal brain. My research group in Children's National Hospital, Washington, DC, has demonstrated that the fetuses with CHD exhibited significantly impaired volumetric and metabolic development of the brain in third trimester (Limperopoulos et al. 2010 Circulation). However, the effects of CHD on fetal brain function and its underlying mechanism remain unclear. In particular, the role of placental insufficiency in CHD-related brain injury has not been well understood.

Our challenge was to find out reliable biomarkers for early diagnosis of fetal brain injuries using noninvasive imaging tools including magnetic resonance imaging (MRI). It was technically challenging since fetal MR images tend to be seriously degraded by motion artifacts (example video). We have developed a robust preprocessing pipeline which addresses fetal motion artifacts and physiological noises in utero (JMI 2016, SPIE 2015, EMBS 2014).

Using the proposed pipeline, we had examined the effects of maternal hyperoxia on the hemodynamic responses of the placenta and fetal brain through functional MRI. In this study, my team first demonstrated that fetal cerebral blood oxygenation was significantly increased during short-term (3-5 minutes) maternal hyperoxia in fetuses with single ventricle physiology of congenital heart disease (Radiology 2019; see highlight). We also have carried out diffusion-weighted imaging (DWI) of the placenta, followed by ex vivo placental dissection for pathology test (see examples), to figure out how placental structure and function are impaired in such high-risk pregnancies as diabetes and IUGR.

AI Empowers Fetal Diagnostic Imaging

Using the state-of-the-arts artificial intelligence (AI) and deep learning technologies, we have actively developed not only robust tools for data processing and analyses (including fetal motion correction and image segmentation) dedicated to the moving fetus (SPIE 2017, SPIE 2017a, ISMRM 2017), but also diagnostic methods for fetal brain impairment for pediatric applications (ISBI 2018, ISPD 2015). We recently developed a deep learning application for fully automatic segmentation of fetal brain and placenta in T2-weighted MR images using convolutional neural networks (ISMRM 2019, ISMRM 2018; see details). We also have developed an integrated software platform, called 'Fetalove', which makes it convenient for doctors and researchers to automatically carry out the entire processes of fetal MRI data processing and statistical analyses.


Software

  • Fetalove: the Fetal MRI Toolbox

Publications

Presentations