基于多分类潜变量SVMs的肺CT中的结节检测

摘要

Two-class latent SVMs and multi-class SVMs based on encoding are combined to design multi-class latent SVMs. The algorithm can fully analyze information of above two SVMs:variety and location of the ob-jects. Multi-class latent SVMs is used to a 65-case database from Jilin Tumor Hospital,and the test results indi-cate that performance of algorithm in the paper is better than other 4 CAD schemes.%结合二分类的潜变量 SVMs 和一种基于编码的多分类 SVMs,设计一种多分类潜变量SVMs,同时具有二者的优点,即考虑检测对象的多样性及位置信息。将多分类潜变量SVMs应用到计算机辅助肺部结节检测,对于来自吉林省肿瘤医院的65组临床病例进行试验,实验结果证明其特异性与灵敏性均优于其他四种当前国际热门的计算机辅助结节检测算法,可以有效辅助放疗师做出最终决策。