Abstract #2412

Combination of Sparse & Wrapper Feature Selection from Multi-Source Data for Accurate Brain Tumor Typing

Vangelis Metsis¹, Ovidiu C. Andronesi^2,3, Heng Huang¹, Michael N. Mindrinos⁴, Laurence G. Rahme⁵, Fillia Makedon¹, Aria A. Tzika^2,3

¹Computer Science & Engineering, University of Texas at Arlington, Arlington, TX, United States; ²NMR Surgical Laboratory, Dept. of Surgery, Harvard Medical School & Massachusetts General Hospital, Boston, MA, United States; ³Athinoula A. Martinos Center of Biomedical Imaging, Department of Radiology, Massachusetts General Hospital, Boston, MA, United States; ⁴Dept. of Biochemistry, Stanford University School of Medicine, Stanford, CA, United States; ⁵Molecular Surgery Laboratory, Dept. of Surgery, Massachusetts General Hospital & Shriners Burn Institute, Harvard Medical School, Boston, MA, United States

In this work we verify the advantage of combining features from Gene Expression and MRS data for brain tumor typing and we introduce a new feature selection method based on Joint ℓ2,1-Norms Minimization which improves classification accuracy in the multiclass problem.