Loukas
G. Astrakas1,2, Konstantinos D. Blekas3, Ovidiu C.
Andronesi1,4, Michael N. Mindrinos5, Peter M. Black6,
Laurence G. Rahme7, A Aria Tzika1,4
1NMR Surgical Laboratory, Department of
Surgery, Massachusetts General Hospital and Shriners Burns Institute, Harvard
Medical School, Boston, MA, United States; 2Department of Medical
Physics, University of Ioannina, Ioannina, Greece; 3Department of
Computer Science, University of Ioannina, Ioannina, Greece; 4Department
of Radiology, Massachusetts General Hospital, Harvard Medical School,
Athinoula A. Martinos Center of Biomedical Imaging, Boston, MA, United
States; 5Stanford Genome Technology Center, Department of
Biochemistry, Stanford University School of Medicine, Palo Alto, CA, United
States; 6Neurosurgery, Brigham and Womens Hospital, Harvard
Medical School, Boston, MA, United States; 7Molecular Surgery
Laboratory, Department of Surgery, Massachusetts General Hospital and
Shriners Burn Institute, Harvard Medical School, Boston, MA, United States
Our
aim was to develop a novel approach that combines high-resolution magic angle
spinning (HRMAS) H1 NMR and genomics in the same biopsies to improve
prognostication of brain tumors. We employed a linear Support Vector Machine
combined with the robust minimum redundancy maximum relevance feature
selection scheme, and applied our algorithm to combined HRMAS 1H MRS and
microarray data of the same adult brain tumor biopsies. Our results
demonstrate that we are able to produce accurate and meaningful data and
introduce a novel classification scheme that predicts a clinically meaningful
parameter such as survival better than either method alone.