兔肝VX2肿瘤是一种快速生长的肿瘤模型, 可以在多种器官如肝、 肺、 直肠等快速生长, 常用于肿瘤研究。 采用可见-近红外高光谱技术对四只兔子的兔肝VX2肿瘤和正常组织进行活体和离体的反射光谱检测, 然后采用支持向量机分别实现了二分类(正常肝组织和肝VX2肿瘤组织)和四分类(未出血活体正常肝组织、 未出血活体VX2肿瘤组织、 出血离体正常肝组织和出血离体肝VX2肿瘤组织)。 根据其光谱反射曲线的特征, 选择了400~1 800 nm区间的数据为特征变量。 为进一步提高分类准确率, 分别采用5折交叉验证和遗传算法对支持向量机的核函数参数 g和惩罚因子 c进行了优化。 其中5折交叉验证优化参数和分类结果为: 二分类优化的惩罚参数 c为4, 核函数参数 g为0.125 0, 其校正集和预测集的准确率都达到了100%; 四分类中优化出的参数 c为8, g为0.121 1, 其校正集和预测集的准确率分别达到了99.242 4%和93.333%。 遗传算法优化参数和结果为: 二分类中优化的参数 c为0.845 6, g为0.062 5, 其校正集和预测集的准确率同样都达到了100%; 四分类中优化的参数 c为5.5307, g为0.068 5, 其校正集和预测集的准确率分别达到了99.242 4%和100%。 结果显示两种优化方法都取得了很好的效果, 遗传算法优化参数对四分类的分类更为精确。 为进一步提升算法速度, 采用间隔选取变量的方法来不断减少特征变量, 最终每隔100 nm谱段选择一个变量, 共选择14个谱段作为特征变量。 采用遗传算法优化支持向量机参数并对其分类进行了研究, 结果表明: 二分类和四分类的校正集和预测集结果准确率均为99.242 4%, 而且运行时间分别为11.4和20.0 s, 与选择全波段的运行时间: 340.3和491.0 s相比, 说明多光谱技术可以进行肝VX2肿瘤组织和正常肝组织的鉴别, 且分类准确率可达99%以上, 而且运行时间缩短了很多。 为未来多光谱技术在未来临床肿瘤诊断中实现肿瘤组织的快速实时在线检测和分类奠定了基础, 显示出巨大的应用潜力。

Rabbit liver VX2 tumor is a tumor model that can grow rapidly in various organs, such as liver, lung, rectum, etc., and is often used in tumor research. In this paper, using high-near-infrared spectrum technology to four rabbits VX2 liver tumor and normal tissue in vivo and in vitro reflection spectrum detection, then respectively the Two categories based on support vector machine (normal liver tissue and liver VX2 tumor tissue) and Four categories (not bleeding living normal liver tissue, not living liver VX2 tumor tissue bleeding, bleeding in vitro normal liver tissue and hemorrhage in vitro liver VX2 tumor tissue). According to its spectral reflection curve characteristics, the data in the range of 400~1 800 nm are selected as characteristic variables. In order to further improve the classification accuracy, the kernel parameter g and penalty factor c of the support vector machine was optimized by using a 50 fold cross-validation and genetic algorithm, respectively. The optimization parameters and classification results of the 50-fold cross-validation are as follows: penalty parameter c of the dichotomy optimization is 4, kernel parameter g is 0.125 0, and the accuracy of the correction set and prediction set reaches 100%. The optimized parameters c and g are 8 and 0.121 1, and the accuracy of the correction set and the prediction set are 99.242 4% and 93.33 3%, respectively. The optimized parameters and results of the genetic algorithm are as follows: the optimized parameters c and g in dichotomy are 0.845 6 and 0.062 5, respectively, and the accuracy of Two categories, the correction set and the prediction set, is agreed to reach 100%.The optimized parameter C in the Four categories was 5.530 7 and g was 0.068 5, and the accuracy of the correction set and the prediction set reached 99.242 4% and 100%, respectively. The results show that the two optimization methods have achieved good results, and the genetic algorithm is more accurate in the classification of the Four categories. In order to further improve the speed of the algorithm, the method of variable selection at intervals was adopted to reduce the characteristic variables continuously. Finally, a variable was selected for every 100 nm spectral segment, and a total of 14 spectral segments were selected as the characteristic variables. Parameters of support vector machine were optimized by using genetic algorithm for the classification was studied, the results show that the Two categories and Four categories of both results of the calibration set and prediction set were 99.242 4%, and the running time of 11.4 s and 20.0 s respectively, and choosing all band running time: 340.3 s and 491.0 s compared to how spectroscopy can be in the identification of hepatic VX2 tumor tissue and normal liver tissue. The classification accuracy rate can reach more than 99%, and the running time shorten a lot. Therefore, it also lays a foundation for realising rapid real-time online detection and classification of tumor tissues in the future clinical tumor diagnosis with multi-spectrum technology, showing great application potential.