Feature selection for the prediction and visualization of brain tumor types using proton magnetic resonance spectroscopy data
Document typePart of book or chapter of book
Rights accessRestricted access - publisher's policy
In cancer diagnosis, classification of the different tumor types is of great importance. An accurate prediction of basic tumor types provides better treatment and may minimize the negative impact of incorrectly targeted toxic or aggressive treatments. Moreover, the correct prediction of cancer types in the brain using non-invasive information –e.g. 1H-MRS data– could avoid patients to suffer collateral problems derived from exploration techniques that require surgery. We present a feature selection algorithm that is specially designed to be used in 1H-MRS (Proton Magnetic Resonance Spectroscopy) data of brain tumors. This algorithm takes advantage of the fact that some metabolic levels may consistently present notorious differences between specific tumor types. We present detailed experimental results using an international dataset in which highly attractive models are obtained. The models are evaluated according to their accuracy, simplicity and medical interpretability. We also explore the influence of redundancy in the modelling process. Our results suggest that a moderate amount of redundant metabolites can actually enhance class-separability and therefore accuracy.
CitationGonzález, F.F.; Belanche, Ll. Feature selection for the prediction and visualization of brain tumor types using proton magnetic resonance spectroscopy data. A: "Computational intelligence methods for bioinformatics and biostatistics". Springer, 2012, p. 83-97.