摘要： 当两类中的样本数量差别较大时，支持向量机的分类能力将会下降。该文提出了一种支持向量机新算法——DFP-PSVM，将有约束条件的二次规划问题转换为无约束二次规划问题，并通过优化计算来实现。为了克服传统的蛇形算法不能收敛于边缘凹陷处以及初始化过于敏感的缺点，采用基于可变形模型的梯度矢量流方法，提取了乳腺X光片中的肿瘤区域，分析了3个基于边缘的价矩。将其他肿瘤形状特征作为DFP-PSVM分类算法的特征输入，进行恶性肿瘤和良性肿瘤的计算机辅助诊断。实验表明，在小样本、两类样本数量“严重不均衡”的情况下，该算法有着较强的分类能力。

关键词: 可变形模型, 梯度矢量流, 肿瘤, 形状特征, 支撑向量机

Abstract: When two-class problem samples are very unbalanced, SVM has a poor performance. A novel SVM algorithm, DFP-SVM is presented to solve the problem implemented by transfering the quadratic program with limited condition into quadratic program without constraining condition. Optimal computation is conducted to achieve exciting results. In order to overcome the problems of traditional snake associated with poor convergence to boundary concavities and sensitive initialization, gradient vector flow based on deformable models is presented to segment tumor region. And three new moments based on boundary are also developed. The novel classifier applies the three moments and other shape features to classify the tumor into the malignant or the benign. Computational results indicate that the modified algorithm has a strong capability of classification for the unbalanced data of small set of samples related to two-class problems.

Key words: deformable model, gradient vector flow, tumor, shape feature, support vector machine

中图分类号: 

• TP273+.5