Classifying Magnetic Resonance Spectra of Brain Neoplasms Using Fuzzy and Robust Gold Standard Adjustments

  • Nicolino Pizzi
Conference paper
Part of the Perspectives in Neural Computing book series (PERSPECT.NEURAL)


Magnetic resonance spectra are often difficult to analyze due to their complex nature and the presence of noise. Many preprocessing methods exist that transform the original input data in order to eliminate or diminish the effects of noise and/or reduce the dimensionality of the input space. Unfortunately, culling diagnostic information is further exasperated by the fact that the reference test or “gold” standard, against which a new and possibly imperfect magnetic resonance diagnostic test is measured, may itself be imprecise or even unreliable. However, little work has been done to investigate a methodology whereby the possible imprecision of a well-established but tarnished gold standard may be addressed while at the same time maintaining its vital discriminatory power.

Two strategies are presented to burnish such tarnished gold standards.

The first uses a fuzzy set theoretic preprocessing method to enhance the gold standard by incorporating non-subjective within-group centroid information.

The second uses a robust estimation of deviations from group medians for the reclassification of spectra in a training set.

These strategies were applied to a multi-layer perceptron classifier using magnetic resonance spectra of human brain neoplasms. The strategies garnered improvements to the classification accuracy of the neural network of 13% and 10%, respectively, as measured using test spectra. Furthermore, the remaining misclassifications were more conservative.


Human Brain Neoplasm Preprocessing Method Test Spectrum Original Input Data Gradient Descent Search 
These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. Klir G. Developments in uncertainty-based information. Adv Computs 1993; 36: 255–332CrossRefGoogle Scholar
  2. Pizzi N, Somoijai, R. Fuzzy encoding as a preprocessing method for artificial neural networks. In: Proceedings of the World Congress on Neural Networks, San Diego, USA, pp 643–648, 1994.Google Scholar
  3. Pizzi N, Somoijai, R. To burnish tarnished gold standards: a classification strategy for MR spectra as applied to human brain neoplasm diagnosis. In: Proceedings of Society of Magnetic Resonance, Nice, France, p 1937, 1995.Google Scholar
  4. Seber G. Multivariate observations. Wiley & Sons, Toronto, 1984.CrossRefMATHGoogle Scholar
  5. Ripley B. Neural networks and related methods for classification. J Royal Stat Soc B 1994; 56: 409–456MATHMathSciNetGoogle Scholar
  6. Everitt B. Analysis of contingency tables, Chapman & Hall, London, 1992.MATHGoogle Scholar

Copyright information

© Springer-Verlag London Limited 1998

Authors and Affiliations

  • Nicolino Pizzi
    • 1
  1. 1.Institute for BiodiagnosticsNational Research Council CanadaWinnipegCanada

Personalised recommendations