Embedding filtering criteria into a wrapper marker selection method for brain tumor classification: an application on metabolic peak area ratios

Abstract

The purpose of this study is to identify reliable sets of metabolic markers that provide accurate classification of complex brain tumors and facilitate the process of clinical diagnosis. Several ratios of metabolites are tested alone or in combination with imaging markers. A wrapper feature selection and classification methodology is studied, employing Fisher's criterion for ranking the markers. The set of extracted markers that express statistical significance is further studied in terms of biological behavior with respect to the brain tumor type and grade. The outcome of this study indicates that the proposed method by exploiting the intrinsic properties of data can actually reveal reliable and biologically relevant sets of metabolic markers, which form an important adjunct toward a more accurate type and grade discrimination of complex brain tumors

Accurate grading of brain gliomas by soft independent modeling of class analogy based on non‐negative matrix factorization of proton magnetic resonance spectra

“…Developing a decision support system for classification of H 1 ‐MRS data is a helpful approach to achieve higher reliability in analysis of multivariate data. Consequently, there has been a significant interest in the development of statistical models (especially multivariate ones) for tumor classification based on 1 H‐MRS data . In some studies, spectral data were coupled with data from imaging studies or were modeled with classifiers such as linear discriminant analysis neural networks or support vector machines (SVMs) for grading of tumor cases .…”

“…In some studies, spectral data were coupled with data from imaging studies or were modeled with classifiers such as linear discriminant analysis neural networks or support vector machines (SVMs) for grading of tumor cases . Some studies applied a feature selection or a robust modeling approach to improve the classifiers . In this study, SIMCA was investigated using both traditional and improved algorithm.…”