Fuzzy preprocessing of gold standards as applied to a neural network classifier of magnetic resonance spectra

Culling diagnostic information from biomedical spectra is often exasperated by an imperfect or imprecise gold standard. A fuzzy set theoretic preprocessing method is described that reduces the classification error rate by enhancing a gold standard through the incorporation of nonsubjective within-group centroid information. Magnetic resonance spectra of human brain neoplasms were used to determine the effectiveness of this strategy. A multi-layer perceptron classifier was used as the performance benchmark