基于CD细胞神经网络的肝脏B超图像数据挖掘

Data dig of liver B-ultrasound image based on counter detection cellular neural networks

目的:运用CD细胞神经网络对肝病患者的肝脏B超图片进行探测,以便更好地在图像中挖掘有用信息,协助临床诊断。方法:实验于2004-07/2005-11在北京西苑医院和北京科技大学完成,实验前期进行图像采集,选取北京西苑医院收治的5例肝病患者的肝脏B超图片用于实验,后期进行数学分析。设计细胞神经网络模板参数的鲁棒性定理,该定理提供了一组参数不等式以确定满足该功能的参数变化区间。利用细胞神经网络方法对肝脏B超图像进行轮廓探测,由此得到肝脏B超图像里面所有轮廓信息,把处理结果数据化,利用十次多项式来进行拟合与数据挖掘,最后通过多项式的系数来确定肝脏的病变情况。结果:按意向处理分析,实验选取5例肝病患者的肝脏B超图片,均进入结果统计。5例肝病患者的最佳逼近系数经治疗后明显趋于健康者的最佳逼近系数。这些数据的变化似乎能表明肝受损的程度,系数a10越大,似乎反映肝损伤越严重;系数a9较小似乎反映肝损伤程度越轻。结论:通过细胞神经网络处理B超图像产生的数据,初步分析似乎提示患者的肝损伤与十次多项式系数之间的某种关系,从而断定肝脏的病变情况,为医生的临床诊断提供参考。

AIM： To detect the liver B-ultrasound image by using the counter detection cellular neural networks （CD CNN）, so as to explore useful information for clinical diagnosis. METHODS： The experiment was conducted in Xiyuan Hospital and University of Science and Technology Beijing from July 2004 to November 2005. Liver B-ultrasound image of five patients with liver disease, . who were treated in Xiyuan Hospital of Beijing, were collected for mathematics analysis. The robust CNN template was designed, which provided a group of parameter inequality to satisfy its parametric change interval. And then the B-ultrasound was detected by CNN to obtain some contour information from the B-scan images, and the result was data processed. Data of the liver will be fitted and mined by polynomials of order ten, and the pathologic status was identified by the coefficient of polynomial. RIESULTS： With intention-to-treat, all 5 liver B-ultrasonnd images entered the result analysis. The best approximation coefficient of 5 patients with liver disease after treatment was obviously equal to the one of healthy people. The change of the data described the injury ,degree of liver：, the bigger the a10 was, the more serious the injury was; the smaller a9 was, the lighter the injury was. CONCLUSION： According to the data that processed the B-scan image with CNN, it seems that there is＇some relation between the damage of liver and coefficients of polynomial, which provides reference for diagnosis of doctors.