A study on MANOVA as an effective feature reduction technique in classification of childhood medulloblastoma and its subtypes

Abstract

Childhood medulloblastoma (MB) is the most common embryo brain tumor and an area that needs utmost attention, as clinical diagnosis can be very difficult in case of infants and children. The rate of survival can increase with prompt diagnosis. Till date, there is no computer aided methodology for identification of childhood medulloblastoma and its subtypes. The diagnosis depends on qualitative visual inspection of the histological slides of the biopsy samples by clinical experts. We convert this qualitative judgment to quantitative features after digitization of the images. The feature set obtained from digital analysis from these biopsy tissues is very large and is computationally expensive. In this study, we examine whether the features are statistically significant for analysis towards classifying childhood MB from normal samples and its various subtypes, using MANOVA. Further, this technique is used as a feature reduction technique, which proves that it can be effectively used as such. Infact, the simplicity of the technique makes it a better choice when considering a sizeably high number of features.