Characterisation of tumorous tissue in rat brain by in vitro magnetic resonance spectroscopy and Artificial Neural Networks

Abstract

The NMR data in this study was interpreted statistically without using molecular and spectroscopical prior knowledge. Only the class origin (tumour/healthy tissue) was used as additional information for the training pattern. The samples were consistently classified successfully. This indicates that sample classification can be automated. Sensitivity analysis of the input pattern can be used for interpretation of NMR spectra without adding any medical or biochemical knowledge. The substances NAA, creatine, myo-inositol, choline and alanine were found to be important, the concentration of lactate seems to be an unreliable parameter for this classification problem. These results are exactly in line with clinical findings and medical interpretation of tumor metabolism. This study demonstrated, that NMR spectroscopy in combination with ANN offers a promising potential for the diagnosis of brain tumors.