摘要
在乳腺肿瘤超声诊断中,医师通过手工缓慢加压形成一个超声图像序列.针对乳腺肿瘤病例的单一图像进行识别,存在被干扰而改变的特征,而且不能反映特征在序列中的变化情况.临床研究也显示,乳腺超声图像序列之间的特征变化是判别其良恶性的一个重要指标.针对这个问题,基于手工加压程度变化的评估,提出了5个新型的基于弹性的归一化特征参数,并且联合形态和纹理特征,综合进行病变分析.SVM分类器实验表明,这种方法能够有效地识别肿瘤良恶性,为医师诊断提供必要的辅助诊断信息.
Breast ultrasound image sequences were usually obtained with a free-hand compression.In a breast ultrasound images classification system based on single image,many features may change due to interferences,and the validity of these features in the sequences can not be judged from a single image.Clinical research has demonstrated that the validity of these features in image sequences is a very important indicator for judging a tumor as benign or malignant.Aiming at the problem,5new normalized elasticity characteristics based on compression depth were proposed in the paper.SVM was applied to classification also.Experimental results show that the proposed method combining texture and morphology characteristics can classify image sequences effectively,providing essential information for making diagnose.
基金
安徽省教育厅自然科学基金重点项目(2006KJ097A)资助
