A feature-pooling and signature-pooling method for feature selection for quantitative image analysis: Application to a radiomics model for survival in Glioma

Abstract

We proposed a pooling-based radiomics feature selection method and showed how it would be applied to the clinical question of predicting one-year survival in 130 patients treated for glioma by radiotherapy. The method combines filter, wrapper and embedded selection in a comprehensive process to identify useful features and build them into a potentially predictive signature. The results showed that non-invasive CT radiomics were able to moderately predict overall survival and predict WHO tumour grade. This study reveals an associative inter-relationship between WHO tumour grade, CT-based radiomics and survival, that could be clinically relevant.