Improving Stability of Feature Selection for Brain Tumour Diagnosis using H-MRS Data

摘要: Magnetic Resonance Spectroscopy for brain tumour diagnosis is progressively replacing harmful biopsy. Nonetheless, dealing with such multidimensional outcome becomes a difficult task for the medical community. Computation-based tools able to effectively reduce dimensionality of data without losing diagnostic ability ease the interpretation of results. The current study presents a novel technique to improve stability of feature subset selection algorithms by means of an instance weighting approach. We report experiments performed on real data showing an improvement on feature selection stability up to 40%.