FastPlex: A Fast Deep Learning Segmentation Method for Accurate Choroid Plexus Morphometry

点击翻译按钮获取中文摘要

PROJECT SUMMARY The choroid plexus (ChP) protrudes into the lumen of the four cerebral ventricles and is the principal source of cerebrospinal fluid (CSF), which together play an important role in neuronal patterning, brain physiology, buoyancy, and maintaining homeostasis by providing physical, enzymatic, and immunological barriers to the brain. Neuroimaging studies have observed ChP morphological changes with aging, neurodevelopmental and neuropsychiatric disorders, which suggests that the ChP may play a role in development and brain disorders. Despite this growing evidence, the ChP has not been the focus of commonly used neuroimaging tools, which causes it to be poorly segmented, mislabeled, and incorrectly quantified. Therefore, there is a critical need to more accurately segment the ChP. The overall objectives for this project are to develop a novel, fast, reliable, generalizable, and dedicated open-source deep learning method for accurate ChP segmentation to understand how the ChP changes across the lifespan and differs among brain disorders. Samples for this study will come from high resolution [Human Connectome Project (HCP) and Connectome Related Human Disease (CRHD)] and conventional (inclusive of high risk for psychosis, first episode and chronic psychosis, bipolar disorder, and Alzheimer’s disease) neuroimaging datasets totaling over 22,000 brains. The rationale for the proposed research is to provide an open-source segmentation tool that will allow for more robust analyses into the ChP’s role in various brain disorders and a better foundational understanding of the how the ChP changes over time with age. To attain the overall objectives, the following three specific aims are proposed: (1) develop and validate a deep- learning method for the accurate segmentation of the ChP; (2) generate ChP volume data across the lifespan that can be used to characterize longitudinal changes and morphological differences across a variety of neuropsychiatric disorders; (3) establish reliability, generalizability, and fairness for broad distribution of FastPlex. To accomplish these aims, a total of 700 brains will be manually segmented – accounting for scanner type and image resolution that is balanced for age, sex, ethnicity/race, socioeconomic status, and brain disorder – to serve as training, validation, and testing labels for the deep-learning tool. Lasty, reliability and generalizability will be established to produce a tool that will be broadly distributed with the research community. The proposed research is innovative and significant because it will focus on an innovative comprehensive ChP segmentation tool (lateral, temporal horn, 3rd, and 4th ventricles) that also estimates partial volume effects and provides super resolution ChP labels, which together will enhance foundational knowledge on ChP neurodevelopmental and neuropsychiatric changes. The results of this research are expected to contribute meaningfully to the understanding of pathologic mechanisms underlying these disorders and to the development of novel strategies targeting specific disease processes.