Prior to her work at NYU, she earned her PhD degree in Biomedical Imaging from Tsinghua University, which included a one year fellowship in the Department of Psychiatry at Yale University. During this time, she developed data-driven methods to measure neural compensation (i.e., the way in which the brain reorganizes itself in older adults to compensate for neural deterioration), and she studied the effects of exercise and cognitive training on brain plasticity over time in older adults.
We are grateful that Dr. Ji was willing to answer a few questions about her work. To read a lay summary of the research paper for which she won this award, please click here. Read on to learn more about Dr. Ji’s research!
Q1: What are the main challenges of conducting fetal MRI studies, and how have you overcome them? Lanxin Ji (LJ): Conducting fetal fMRI research comes with its unique set of challenges that distinguish it from adult imaging. In a typical fMRI scan, subjects need to remain still for a long period (typically over 10 minutes) to obtain high-quality data. However, this isn't possible when working with fetuses. Fetal movements, both unpredictable and unavoidable, can disrupt the fetal MRI data. So, managing this motion is a significant challenge. In addition, the data collection method also differs for fetal MRI. While adult MRI scans usually employ head coils to gather data directly from the brain, with fetuses, we use abdominal coils placed on the mother's belly. Fetal brains are surrounded by maternal tissues, creating a more complex background compared to the relatively clean background in adult MRI. These distinctions result that established data processing pipelines designed for adults may not be readily applicable to fetal fMRI.
While these challenges are significant, our lab—led by Dr. Moriah Thomason, a pioneer in fetal brain imaging—has been at the forefront of overcoming them by developing methodologies for optimally acquiring and processing fetal fMRI. In this work on “Fetal behavior during MRI changes with age and relates to network dynamics”, we manually selected data segments with minimal motion to estimate neural signals and applied individual-level ICA denoising to mitigate motion-related artifacts. Additionally, we employed deep learning techniques to separate fetal brains from maternal tissues. We also used optimal strategies for various preprocessing steps, including normalization, masking, denoising, and smoothing, using our previous evaluation of their relative impacts to make informed decisions in our data processing workflow. Q2: What are the main take-aways of this work? LJ: At a high-level, our study demonstrates the groundbreaking capability to simultaneously capture fetal behavior and neural activity through fetal fMRI. We discovered that natural movements of fetuses during fMRI scans, which have long been considered barriers to effective imaging, can be quantified as meaningful indicators of fetal behavior. Our findings provide the initial evidence of neural activity patterns being linked to both fetal motion at the time of measurement as well as future infant motor behavior. Specifically, for more mature fetuses or those displaying more movements, their motor cortex tends to exhibit lower connectivity to brain regions within its community, possibly reflecting a maturation process in which the motor network extends its connections to other large-scale networks.
Q3: What is required to bring fetal neuroimaging to the next level? LJ: Advanced data analysis methods are needed to extract meaningful information from fetal neuroimaging data. We know many research groups, including us, are working to establish best practices for collecting and processing fetal neuroimaging data from various angles. This involves refining motion correction algorithms, addressing magnetic field inhomogeneities, reconstructing data disrupted by motion, creating normalization algorithms tailored to the contrast of fetal images, and designing denoising strategies specifically for fetal data. As these efforts continue to advance, they hold the potential of establishing a consensus in the field of fetal neuroimaging, which will greatly enhance the reproducibility of findings and promote comparability across studies. Further, the integration of data from multiple imaging modalities, such as fMRI, DTI, and structural MRI, will provide a more comprehensive view of fetal brain development. In addition, larger and diverse open datasets are also crucial in drawing reliable conclusions and furthering clinical applications.