基于BP算法的前馈神经网络超声诊断方法的研究

Study on ultrasound diagnostic method of feedforward neural network on the basis of BP algorithm

目的:研制一套提升超声诊断水平的神经网络超声诊断软件系统,以提高超声诊断效率和降低诊断误判率。方法:基于神经网络理论设计误差反向传播(BP)算法的前馈神经网络结构,通过数据库累积超声专家的诊断知识对超声诊断图像进行特征提取,利用完成分类的超声图像数据,训练神经网络以获取网络权重,将目标超声图像导入神经网络得到诊断分类。结果:将患者肝部超声图像导入系统,在病理特征不明显情况下,诊断系统仍可快速做出诊断分类,并给出置信区间,辅助医师进行诊断。超声辅助诊断系统的诊断时间由通常的10 min左右降低至2 min左右,缩短了患者的就诊时间。结论:通过临床测试,系统不仅可以提高超声诊断效率、降低就诊时间,而且可通过专家知识库实现专家知识共享,提高患者疾病诊断的准确率。

Objective:To develop a software system of ultrasound diagnosis of neural network that could improve the ultrasound diagnosis level,so as to enhance diagnostic efficiency and reduce rate of misdiagnosis.Methods:Firstly,the structure of feedforward neural network of error back propagation(BP)athigrathm was designed on the basis of neutral network theory.Secondly,feature extractions for images of ultrasound diagnosis were implemented through the diagnostic knowledge of accumulation ultrasound specialists of data base.Thirdly,the data of ultrasonic images that were used to complete classification were used to train neural network so as to obtain weight of network,and to import ultrasonoscopy of targets in neural network so as to obtain diagnostic classification.Results:After the color ultrasonic images of patients’liver were imported into system,the diagnostic system still could quickly make the diagnosis classification,and provide confidence interval and implement diagnosis to assist physicians under the conditions that pathological characteristics were not obvious.And the diagnostic time of ultrasonically auxilliary diagnosis system were reduced from routine 10 min to 2 min,and it shortened the clinical time of patients.Conclusion:The results of clinical test indicate that the system can not only improve the efficiency of ultrasonic diagnosis and reduce the clinical time,but also can realize sharing of experts’knowledge and improve the diagnostic accuracy rate for disease of patients through expert knowledge base.