Improving the classification of brain tumors in mice with perturbation enhanced (PE)-MRSI

摘要: Classifiers based on statistical pattern recognition analysis of MRSI data are becoming important tools for the non-invasive diagnosis human brain tumors. Here we investigate potential interest perturbation-enhanced (PE-MRSI), in this case acute hyperglycemia, improving discrimination between mouse MRS patterns glioblastoma multiforme (GBM), oligodendroglioma (ODG), and non-tumor parenchyma (NT). Six GBM-bearing mice three ODG-bearing were scanned at 7 Tesla by PRESS-MRSI with 12 136 ms echo-time, during euglycemia (Eug) also induced hyperglycemia (Hyp), generating altogether four datasets per animal (echo time + glycemic condition): 12Eug, 136Eug, 12Hyp, 136Hyp. For classifier development all spectral vectors (spv) selected from matrix unit length normalized (UL2) used either as a training set (76 GBM spv, mice; 70 ODG two 54 NT spv) or an independent testing (61 31 ODG, one mouse; 23 spv). All Fisher's LDA classifiers obtained evaluated far their descriptive performance—correctly classified cases (bootstrapping)—and predictive accuracy—balanced error rate classification. MRSI-based 12Hyp consistently more efficient separating GBM, regions, overall accuracies always >80% up to 95–96%; remaining within 48–85% range. This was confirmed user-independent selection sets, using leave-one-out (LOO). highlights protocols characterization preclinical