Meeting Banner
Abstract #3355

Spectral Prototype Extraction for the Discrimination of Glioblastomas from Metastases in a SV 1H-MRS Brain Tumour Database

Sandra Ortega-Martorell1,2, Ivn Olier3, Alfredo Vellido4, Margarida Juli-Sap2,5, Carles Ars1,2

1Grup dAplicacions Biomdiques de la RMN (GABRMN), Departament de Bioqumica i Biologia Molecular, Universitat Autnoma de Barcelona (UAB), Cerdanyola del Valls, Barcelona, Spain; 2Centro de Investigacin Biomdica en Red en Bioingeniera, Biomateriales y Nanomedicina (CIBER-BBN), Zaragoza, Spain; 3Institut de Neurocincies (INc), Universitat Autnoma de Barcelona (UAB), Cerdanyola del Valls, Barcelona, Spain; 4Departament de Llenguatges i Sistemes Informtics (LSI), Universitat Politcnica de Catalunya (UPC), Barcelona, Spain; 5Grup dAplicacions Biomdiques de la RMN (GABRMN), Departament de Bioqumica i Biologia Molecular , Universitat Autnoma de Barcelona (UAB), Cerdanyola del Valls, Barcelona, Spain


Spectral Prototype Extraction (SPE) is a novel feature extraction technique, offering the capability of creating spectral prototypes which correspond to known metabolites or groups of metabolites. Experimental results for discriminating glioblastomas from metastases in a SV 1H-MRS brain tumour database, for long and short echo times shows the following: 1) Most of the prototypes describing the dataset can be used as readily interpretable input features in classifiers and 2) SPE-based classification yields results that are comparable to those of PCA-based classification.