MR Spectroscopy

In collaboration with Hyunsuk Shim from Emory, our team is developing new methods for processing 3D whole-brain MR spectroscopy (SMRI) data to enable clinical translation and to make SMRI the standard of care for gliomas. SMRI generates a spectral profile at each imaging voxel, and the metabolite concentrations can be used detect non-enhancing neoplastic infiltration that is not detectible on standard T2 imaging. SMRI can enable better sampling of malignant regions during biopsy, and better treatment of non-enhancing areas that lead to early recurrence.

Our group has developed new spectral fitting algorithms based on convolutional networks to reduce post-imaging processing times from over an hour to several minutes to meet the requirements for routine clinical use and to improve fitting quality. We have also developed novel artifact filtering algorithms based on supervised learning to identify voxels with poor acquisition quality to enable whole-brain imaging and to aid neuroradiologists who lack specialized MRSI expertise. Together with work on new imaging sequences and visualization software, these improved data processing pipelines are supporting clinical trials using SMRI to make treatment-altering decisions.