Abstract #1583

Du S, Sajda P, Mao X, Shungu D

Advanced MR hardware and spectroscopic imaging techniques have enabled dramatic improvements in the spatio-temporal resolution of MRSI data acquisition. However such gains bring in an undesired side effect: high dimensional spectroscopic imaging datasets which challenge available analysis methods. Recently, we introduced cNMF, a new data analysis method for recovering physically meaningful spectral signatures by decomposing the observed spectral data into two matrices. In this study we demonstrate that by applying cNMF to multislice 1H MRSI data of two neurodegenerative diseases affecting children : Batten disease and MELAS, whose phenotypic and radiologic similarities make it difficult to differentiate between them without extensive and costly genetic analyses, we can both extract their respective spectral patterns as well as the spatial distributions of lesion patterns critical for differentiation.