Development of automated web-based spectroscopic MRI clinical interface

点击翻译按钮获取中文摘要

Glioblastoma (GBM) is the most common adult primary brain tumor and is highly aggressive in its disease course. Infiltration of glioma cells into surrounding normal brain make curative surgical resection of GBM impossible, and almost all will eventually recur. Thus, extraordinary significance is placed on radiation therapy (RT) strategies, which have been shown to be effective, but require strong imaging evidence to guide RT planning. Currently employed clinical imaging modalities include T1-weighted contrast-enhanced (CE) MRI, which only identifies leaky neovasculature associated with high grade tumor, and T2-weighted MRI, which is not specific for tumor infiltration. Through advances in neurosurgery, it is now possible to achieve complete or near-complete resection of the CE tumor component in many cases; thus, the region that is treated with the highest RT dose is limited to the empty resection cavity plus a small margin. Due to the generally larger size of the T2 area and unknown status of disease, it is treated to a lesser “microscopic disease” dose. Many times, however, this microscopic disease dose is inadequate to control the tumor. Spectroscopic MR imaging (sMRI) provides a highly sensitive and specific means of identifying these regions, although sMRI has not yet seen use in RT planning due to a lack of clinical decision support software for the analysis, display, and management of sMRI data. Three key hurdles to be overcome are: 1) lack of an automatic, fast and reliable method for spectral quality control; 2) the necessity of quantification of metabolite levels relative to a patient's baseline metabolism; and 3) a clinician- friendly display of the sMRI spectra encoded as a high-resolution, continuous, 3D image set for direct registration and incorporation into the RT planning process. Currently, sMRI processing requires skilled user intervention and shepherding data between several tools, resulting in a complex workflow that takes hours and is impractical for routine use in a fast-paced clinical RT environment. To automate this pipeline and provide clinically useful information to radiation oncologists, we seek to develop a software framework for the automated and expedient processing of sMRI for use in RT planning. We will use novel advances in the fields of high performance computing and deep learning. Specifically, we will develop algorithmic filters for identifying (and eliminating) spectral artifacts, algorithms for personalized localization of tumor infiltration, and methods and interfaces for the volumetric display of sMRI data needed for RT planning and review. Success in the proposed work will produce an automated “scanner-to-clinician” platform for quantitative, expedient, and objective analysis methods to integrate sMRI into routine clinical applications. This tool will also be highly valuable in the MRS-based diagnosis and evaluation of numerous other neuropathologies, including other primary (and metastatic) brain tumors, stroke, multiple sclerosis (and other demyelinating diseases), inborn errors of metabolism, and neurodegenerative diseases.