Comparative Study of Three Neural Approaches in Class Prediction of Cancer

Abstract

Accurate diagnosis and classification is the key issue for the optimal treatment of cancer patients. Several studies demonstrate that cancer classification can be estimated with high accuracy, sensitivity and specificity from microarray-based gene expression profiling using artificial neural networks (ANN). In this paper, a comprehensive study was undertaken to investigate the potential value of other neural networks for the discrimination of acute lymphoblastic leukemia (ALL) and acute myeloid leukemia (AML). Probabilistic neural networks (PNN), multilayer perceptrons (MLP) and the learning vector quantization network (LVQ) were applied for this purpose. The best results were obtained by PNN, followed by MLP networks and LVQ. PNN classifier yields 100% recognition accuracy and is well suited for the AAL/AML classification in cancer treatment. This study presents the capabilities of PNN, and also indicates that PNN should be evaluated in a larger prospective study. Our future work will focus on applying the gene selection method and the PNN network on other dataset to observe the generality of this strategy.