中文摘要：

乳腺钼靶摄影（mammography）是目前诊断乳腺癌的一种重要手段，肿块是钼靶图像中表征乳腺病变的主要病灶之一。文中提出了一种基于改进型Multi—Agent（MA）多分类器融合算法，用来对乳腺肿块的良恶性进行分类。它在经典MA算法思想的基础上，对各单分类器的分类效果、识别性能有所差别的情况进行针对性修正；同时在统计训练数据时，不再依靠分类结果的标签，而是用统计分类结果的置信度来代替；而当算法的迭代训练超过一定次数仍未能趋向稳定时，采用概率加权平均代替原来的简单平均进行整体决策，并用整体溯潦情况调整决策结果，从而帮助算法收敛。此方法在公共数据库DDSM上做了测试，实验结果表明，在单视角下，改进型MA融合算法的分类精度达到了95．63％，优于经典的MA融合算法，其稳定性也较大多数单分类器以及多分类器融合算法要好；在双视角下，改进型MA融合算法的分类精度达到了97．79％，相较其他分类方式也具有一定的优势。

英文摘要：

Mammography is the first choice for diagnosing breast cancer at present, and mammographic tumour is one of the most important symptoms of breast cancer. This paper proposes a multi-classifier fusion scheme with an improved Multi-Agent （MA） algorithm to identify the benign and malignant masses. Based on the MA algorithm, the situation that individual classifiers have different classification and identification performances is modified intentionally. Furthermore, when the improved MA algorithm counts the classification result, not the classification label but the confidence is utilized. When the number of iterations achieves a threshold and the result still has not reached to stabilization, the probability-weighted average algorithm is adopted to replace simple averaging to carry out overall decision-making for the multi-classifier fusion, and the decision-making result is adjusted with the agent tracing situa- tion. The approach was tested on broadly used public database of mammography （DDSM）. Experimental result indicates that under the single-view, the classification precision of the improved MA algorithm achieves 95.63% , which is better than that of the traditional MA fusion algorithm, and its stability is also better than those of most single-classifier and multi-classifier fusion algorithms. Under the multi-views, the classification precision of the improved MA algorithm reaches 97.79% , which is also an advantage compared with other classification algorithms.