Alexandra Constantin1, Adam Elkhaled2, Trey Jalbert2, Radhika Srinivasan2, Soonmee Cha3, Susan M. Chang3, Ruzena Bajcsy1, Sarah J. Nelson2
1Electrical Engineering & Computer Science, University of California, Berkeley, CA, United States; 2Department of Radiology & Biomedical Imaging, University of California, San Francisco, CA, United States; 3Department of Neurological Surgery, University of California, San Franisco, CA, United States
In this study, we applied multivariate pattern recognition methods to HRMAS spectra from image guided tissue samples in order to identify metabolites that are predictive of malignant transformations in gliomas and to accurately detect those patients exhibiting malignant transformations. Our method extracts a small subset of features in the HRMAS spectra and uses it to build a parsimonious model capable of discriminating between patients with different tumor grades with over 90% accuracy. The features used in our model are traced back to known metabolites in the corresponding chemical shift range, thus identifying a useful set of metabolites to acquire in-vivo.