基于分布先验的半监督FCM的肺结节分类

肺结节的良恶性分类对于肺癌的早期发现及诊断具有重要意义。然而实际应用中，标记的图像数量较少，且获取标记将耗费大量的人力，在这种情况下，使用半监督学习算法是有效提高分类性能的一个思路。作为一种经典的半监督学习算法，传统的半监督FCM在未标记样本与标记样本分布不平衡情况下不能充分利用标记信息。针对此问题，本文提出了一种基于分布先验的半监督FCM算法。首先计算样本的先验分布概率，基于获得的先验概率，给样本赋予权重，并将其融入到半监督FCM聚类中，从而强化少量的标记样本在聚类过程中的指导作用。文中在LIDC数据库上进行了相应的实验，实验结果证明，相比较传统的半监督FCM算法，提出的算法能够取得更好的肺结节分类性能。

Classification of pulmonary nodules by semi-supervised FCM based on prior distribution

The classification of pulmonary nodules is significant for the early detection and treatment of lung cancer. However, in real clinical applications, few medical images are labeled and it is difficult to obtain these labels. Semi-supervised learning methods that utilize supervised information to label images may be employed to improve the classification performance of pulmonary nodules. Traditional semi-supervised methods ignore the use of label information when the distribution between labeled and unlabeled specimens is imbalanced. To solve this problem, we propose a semi-supervised fuzzy c-means (FCM) algorithm based on prior distribution for classifying pulmonary nodules. This algorithm first calculates the prior probability of the specimens. Based on the obtained probability, a weight is assigned to each specimen for clustering to intensify the instruction role of a few labeled specimens in the clustering process. We conducted a corresponding test for Lung Image Database Consortium (LIDC) database. The result shows that, compared with the traditional semi-supervised FCM algorithm, the proposed algorithm can obtain better classification performances of pulmonary nodules.