Spectral Separation Analyses of Proton MRSI Data: Validation with Tumor Grade of Brain Glioma
Parra L, Du S, Huang W, Karimi S, Thakur S, Su Y, Sajda P
City University of New York
A constrained non-negative matrix factorization (NMF) algorithm was applied to analyze clinical proton MRSI data of brain gliomas (n = 14), extracting constituent spectra of tumor tissue, normal brain tissue, and residuals. The Cho-NAA patterns of the constituent tumor spectra were consistent with the pathologically proven tumor grades. This MRSI data analysis method resolves partial volume effect, reduces variability of in vivo spectra, and has potential in clinical applications, such as defining tumor margins, treatment planning of radiotherapy, and surgical decisions.