主成分分析结合感知器在医学光谱分类中的应用

基于光谱分析技术的经典理论,应用主成分分析方法对83例(癌症42,非癌41)乳腺患者病理片的紫外吸收光谱进行主成分提取.选用其中44例(癌症22,非癌22)作为训练样本,其余的39例(癌症20,非癌19)作为测试样本,将其主成分数据作为输入向量分别对离散型和连续型感知器神经网络模型进行训练和测试.通过对比发现:离散型感知器模型由于其输出函数值只有{0,1}且算法较为简单,其癌症识别率只有43.3%,非癌识别率为38.7%;而连续型感知器模型将模糊集合理论引进了神经网络系统,将二值{0,1)扩展到隶属度函数的单位区间0,1]上,结果表明这种模型的癌症识别率为83.6%,非癌的识别率为76.3%,取得了较为理想的识别效果.

Principal Components Analysis with Sensation Network Applied in the Recognition of Medicine Spectrum

On the basis of classical theory about spectral analysis, the present article used the method of principal component analysis to get the specificity of 83 ultraviolet absorption spectra from mammary gland patient pathology pieces of 83 cases. The authors chose 44 principal component data as training samples and the rest 39 as testing samples. After training discrete and continual sensation network, the authors found that the recognition rate of cancer was only 43.3% and the recognition of noncancerous one was 38.7% when using the discrete sensation network. However, because fuzzy-mathematics was introduced to the continual sensation network and the output value of this model was expanded to 0,1], the recognition rate of cancer reached 83.6% and that of noncancerous one was 76.3% when using this model.