Automated diagnosis of brain tumours using a novel density estimation method for image segmentation and independent component analysis combined with support vector machines for image classification

Abstract

A computer-aided system was developed for the automatic diagnosis of brain tumours using a novel density estimation method for image segmentation and independent component analysis (ICA) combined with Support Vector Machines (SVM) for image classification. Images from 87 tumor biopsies were digitized and classified into low and high-grade. Segmentation was performed utilizing a density estimation clustering method that isolated nuclei from background. Nuclear features were quantified to encode tumour malignancy. 46 cases were used to construct the SVM classifier. ICA determined the most important feature combination. Classifier performance was evaluated using the leave-one-out method. 41 cases collected from a different hospital were used to validate the systems' generalization. For the training set the SVM classifier gave 84.9%. For the validation set classification performance was 82.9%. The proposed methodology is a dynamic new alternative to computer-aided diagnosis of brain tumours malignancy since it combines robust segmentation and high effective classification algorithm. © Springer-Verlag Berlin Heidelberg 2004.