基于贝叶斯粗糙集的肺部肿瘤CT图像高维特征选择算法

An algorithm for high dimension feature selection of lung tumor CT image based on Bayesian rough set

针对变精度粗糙集在高维特征选择过程中对分类错误率β的过分依赖问题,结合遗传算法提出一种基于贝叶斯粗糙集的肺部肿瘤CT图像高维特征选择算法。首先提取3000例肺部肿瘤CT图像ROI区域的104维特征构造决策信息表;然后从全局相对增益函数的角度分析属性重要度,结合属性约简长度、基因编码权值函数三者的加权和构造一个适应度函数框架,提出以此为启发式信息的属性约简算法;最后利用支持向量机进行分类识别。实验结果表明,本研究算法摆脱了阈值人工设置的束缚,并且在很大程度上提高整体性能,对肺部肿瘤计算机辅助诊断具有积极的推广价值。

Aiming at the over-dependence problem of variable precision rough set in the high dimensional feature selection process,to propose an algorithm for high dimension feature selection of lung tumor CT image based on Bayesian rough set combined with genetic algorithm.Firstly,the 104 dimension features were extracted to construct a decision information table based on 3000 CT ROI regions of lung tumor images.Then,a general fitness function framework was constructed based on weighted sum of attribute reduction length,gene coding weight function and the attribute importance degree of global relative gain function.At the same time,an attribute reduction algorithm was proposed based on these heuristic information.Finally,support vector machine was used to classify the reduction subset and multiple indicators were used to evaluate the model.Experimental results show that the algorithm breaks away from the restriction of threshold artificial settings compared with variable precision rough set,and improves the global performance of the model to a large extent,which has a positive influence for the computer aided diagnosis of lung tumors.