Fetal MRI: robust self-driving brain acquisition and body movement quantification

点击翻译按钮获取中文摘要

PROJECT SUMMARY/ ABSTRACT Our premise is that the fetal stage of human brain development is the most dynamic, the most vulnerable and the most important for lifelong behavioral and cognitive function. As many neurological disorders have their genesis in fetal life, there is a need to accurately quantify normal and abnormal fetal brain development from both the perspective of fetal brain structure and body motion. Better imaging tools would enable us to explore how fetal neurological disorders as well as environmental exposures, such as opioids, maternal obesity, or COVID-19, impact early brain structure and body movements. Magnetic resonance imaging (MRI) T2-weighted, single-shot fast-spin-echo (e.g. HASTE) images provide a unique window into this critical phase of structural brain development, with the potential to detect subtle abnormalities. However, fetal brain MRI is challenging due to fetal motion, which leads to image artifacts, double oblique acquisitions and incomplete brain coverage. As a result, trained MR technologists must “chase the fetus” to amass the necessary images to diagnose the presence or absence of lesions, resulting in long scan times and higher RF energy deposition. Thus, fetal brain MRI is inefficient, limited to specialized centers, and diagnosis is still difficult because fetal motion results in each image being an independent slice that cannot be referenced to another slice, making confirmation of suspicious findings difficult. At the same time, fetal motion is an important measure of functional neurological integrity, informing postnatal outcomes. However, current clinical MR and ultrasound assessments of fetal motion do not fully capture the complex 3D motions of all body parts simultaneously. Better assessment of fetal neurological health requires novel tools to automatically and efficiently obtain coherent, high quality HASTE fetal brain volumes and to characterize 3D fetal whole-body motion. To address these unmet needs, we will leverage convolutional neural network (CNN) models and propose the following aims: (1) Develop a self-driving engine for efficient acquisition of high-quality HASTE fetal brain volumes and (2) Enable automated fetal whole-body motion tracking and characterization. We will deploy the proposed tools in a prospective study that compares fetuses with Chiari II malformation (spina bifida), a disorder known to have brain abnormalities and often associated with decreased leg movement, to typical fetuses with the following aim: (3) Assess performance of the self-driving HASTE engine and whole-body motion characterization in Chiari II vs typical fetuses. For Aims 1 and 2, we will include data from collaborating sites and strategies for CNN generalization to increase robustness and potential to deploy our tools to other scanners. The ability to automatically obtain high-quality coherent fetal brain volumes and characterize fetal motion will improve stratification for fetal treatments and characterization of response to fetal interventions. Success will also enable sites without fetal imaging experts to locally assess and triage fetuses with suspected abnormalities to specialized treatment centers, as well as facilitate large population-based studies to understand the impact of environmental influences on early brain development and fetal behavior.