Data and feature mixed ensemble based extreme learning machine for medical object detection and segmentation

Abstract

Extreme learning machine (ELM) is a single-hidden layer feed-forward neural network with an efficient learning algorithm. Conventionally an ELM is trained using all the data based on the least square solution, and thus it may suffer from overfitting. In this paper, we present a new method of data and feature mixed ensemble based extreme learning machine (DFEN-ELM). DFEN-ELM combines data ensemble and feature subspace ensemble to tackle the overfitting problem and it takes advantage of the fast speed of ELM when building ensembles of classifiers. Both one-class and two-class ensemble based ELM have been studied. Experiments were conducted on computed tomography (CT) data for liver tumor detection and segmentation as well as magnetic resonance imaging (MRI) data for rodent brain segmentation. To improve the ensembles with new training data, sequential kernel learning is adopted further in the experiments on CT data for speedy retraining and iteratively enhancing the image segmentation performance. Experiment results on different testing cases and various testing datasets demonstrate that DFEN-ELM is a robust and efficient algorithm for medical object detection and segmentation.