用于癌症分子分型的虚拟可重构结构演化硬件

Virtual reconfigurable architecture-based evolvable hardware for molecular classification of cancers

根据DNA微阵列数据维数高、样本少的特点,提出了一种用于癌症分子分型的演化硬件方法.该方法是基于虚拟可重构结构的演化硬件识别方法,具备灵活、高速和自适应能力强的特点,有利于建立一个拥有高数据吞吐能力、学习结果易读的高效分类系统.采用信噪比的方法进行特征选择,对选择的基因经过归一化和二值化,然后用演化硬件识别系统通过系统学习和系统分类2个阶段进行处理.急性白血病分类的硬件实验结果表明:演化硬件的识别率和识别时间分别达到了95.88%和0.12μs.

DNA microarray datasets are characterized by high-dimension and limited number of sampies. Thus, an evolvable hardware （EHW） was proposed for the molecular classification of cancers. The virtual reconfigurable architecture-based EHW method has flexibility, high processing speed and self-adaptation ability. These features are useful to build an efficient classification system which provides the high data throughput and the readability of the learned results. For the classification of acute leukemia, the feature selection was based on the signal-to-noise ratio method and the selected normal- ized genes were processed by the EHW classification system through the phases of system learning and system classification. Hardware experimental result shows that the recognition rate and recognition time of the proposed evolvable system for the classification of acute leukemia dataset are 95.88 % and 0.12 μs, respectively.