提高结肠早癌诊断率的荧光光谱小波特征提取与神经网络分类方法研究

提出了一种新的荧光光谱特征提取与模式分类方法以提高激光诱导5-ALA-PpIX荧光光谱对早期结肠癌的诊断准确率。利用小波多尺度分析法提取荧光光谱特征量，对提取的特征量采用基于Rprop算法的BP神经网络进行模式分类。对20只DMH诱导的SD大鼠，12只诱导鼠的第二代鼠，8只正常SD大鼠，按25 mg·kg-1体重剂量尾静脉注射5-ALA溶液，150 min后利用波长为370 nm的钛宝石激光进行活体检测。对预处理后的504条荧光光谱，利用小波多尺度分析法提取12个特征量，BP神经网络将其分为正常组与非正常组(非典型增生、早癌和进展期癌)，分类敏感性与特异性分别为98.91%和97.2%，非典型增生、早癌及进展期癌对正常组织的识别准确率分别为91.3%, 98.85%及98.79%。结果表明此方法不仅对早期结肠癌具有较高的诊断率，且更利于临床实时诊断。

Study of the Methods of Wavelet Feature Extraction and Neural Network Classification of Fluorescence Spectra to Improve the Diagnostic Rate of Colonic Earlier Stage Cancer

In order to improve the diagnostic rate of earlier stage colonic cancer with laser-induced 5-ALA-Pp IX fluorescence spectra, a novel method of extraction of fluorescence spectral feature using wavelet analysis and classification using artificial neural network trained with resilient back-propagation algorithm (R-BPNN) was developed. 504 spectra were collected from 8 normal SD rats, and 20 1,2-DMH-induced SD colon cancer models and 12 second generation rats of induced rats. 150 min later trail intravenous injections of 5-ALA dose of 25 mg x kg(-1) body weight (BW), and fluorescence spectra excited with 370 nm Ti-laser were collected in vivo. After preprocessing, 12 feature variants were extracted with wavelet analysis. With R-BPNN, all spectra were classified into two categories: normal or abnormal, which included dysplasia, early carcinoma (EC) and advanced carcinoma (AC). The sensitivity and specificity were 98.91% and 97.2% respectively. The accuracy of discriminating dysplasia, early carcinoma, and advanced carcinoma from normal tissue were 91.3%, 98.9% and 98.8 respectively. The result indicated that this method could effectively and easily diagnoses earlier stage colonic carcinomas.