A virtual reconfigurable architecture-based intrinsic evolvable hardware (EHW) is proposed for the molecular classification of cancer. To efficiently process DNA microarray datasets and cooperate with the hardware realization of EHW, five different filter-based gene selection methods are compared and discussed in this paper. The EHW classification system handles the selected informative genes through two stages: system
learning and system classification. Empirical studies on a human acute leukemia dataset demonstrate that classification accuracy of the gene selection scheme based on signal-to-noise ratio outperforms its competitors. Classification accuracy of the proposed EHW is high comparable with other state-of-the-art pattern recognition methods. The system recognition time is reduced to 0.12 ms.