Neural Network Ensemble Based on Feature Selection for Non-Invasive Recognition of Liver Fibrosis Stage

Contemporary medicine concentrates on providing high quality diagnostic services, yet it should not be forgotten that the comfort of the patient during the examination is also of high importance. Therefore non-invasive methods that allows to precisely predict the state of the disease are currently one of the key issues in the medical business. The paper presents a novel ensemble of neural networks applied to recognition of liver fibrosis stage from indirect examination method. Several neural network models are build on the basis of outputs of different feature selection algorithms. Then an ensemble pruning procedure with the usage of diversity measures is conducted in order to eliminate redundant predictors from the pool. Finally the weights of classifiers in the fusion process are assessed to establish their influence on the output of the whole ensemble. Proposed method is compared with several state-of-the-art ensemble methods. Extensive experimental investigations, carried out on a dataset collected by authors, show that the proposed method achieve a satisfactory level of the fibrosis level recognition, outperforming other machine learning algorithms and thus may be used as a real-time medical decision support system for this task.