论文部分内容阅读
卵巢癌是目前死亡率最高的妇科疾病之一,而如果得到早期诊断和治疗,卵巢癌患者的存活率可达90%。针对卵巢癌早期诊断问题,基于卵巢癌磷脂质类数据,提出了一种结合缠绕法和过滤法、按照诊断类别相关度挑选特征,然后依据特征标志物的分类率稳定度高低,提取用于诊断早期卵巢癌的特征子集的策略。该方法克服了分类率监督方法忽略生物相关性、依赖分类器易产生过拟合的不足,同时保持了较高的分类率。实验表明,该方法挑选的特征标志物包含更多的分类信息,其分类正确率达到88.9%,且比经典的分类率监督方法和差异表达方法在稳定性能上存在优势。此外,提出的新的标幺化方法去掉了批次差异,获得更好的分类效果,且所选的特征标志物得到生物学关联意义上的支持,具有较高的可信度和实用性。
Ovarian cancer is currently one of the highest mortality of gynecological diseases, and if early diagnosis and treatment, ovarian cancer patients survival rate of up to 90%. Aiming at the early diagnosis of ovarian cancer, based on the phospholipid data of ovarian cancer, a combination of winding method and filtration method was proposed to select the features according to the diagnostic relevance of the categories. Then, based on the classification rate stability of the characteristic markers, Strategy for the subset of characteristics of early ovarian cancer. This method overcomes the shortcomings of the classification rate supervising method that ignore the biological relativity, relies on the classifier easy to over-fit, while maintaining a high classification rate. Experiments show that this method has more classification information, and the classification accuracy rate is 88.9%. And it has more stability than classical supervised classification method and differential expression method. In addition, the new per unitary method eliminates batch differences and obtains better classification results, and the selected characteristic markers are supported by biological relevance, which has high credibility and practicability.