基于神经网络的超声背散射零差K成像脂肪肝评价方法研究  

Hepatic steatosis assessment method using ultrasound backscatter homodyned-K imaging based on neural network estimator

作　　者：丁琪瑛[1] 吴水才[2] 崔博翔 周著黄[2] DING Qi-ying;WU Shui-cai;TSUI Po-hsiang;ZHOU Zhu-huang(Department of Ultrasound,BJUT Hospital,Beijing University of Technology,Beijing 100124,China;Faculty of Environment and Life,Beijing University of Technology,Beijing International Science and Technology Cooperation Base for Intelligent Physiological Measurement and Clinical Transformation,Beijing 100124,China;College of Medicine,Chang Gung University,Taoyuan 33302,Taiwan Province,China)

机构地区：[1]北京工业大学校医院超声室,北京100124 [2]北京工业大学环境与生命学部,智能化生理测量与临床转化北京市国际科研合作基地,北京100124 [3]台湾长庚大学医学院,中国台湾桃园33302

出　　处：《医疗卫生装备》2023年第1期19-26,共8页Chinese Medical Equipment Journal

基　　金：北京市自然科学基金项目(4222001);国家自然科学基金项目(11804013,61871005)。

摘　　要：目的 :针对传统超声背散射零差K模型参数估计方法存在复杂度较高的问题,提出一种基于神经网络的超声背散射零差K模型参数估计方法,并在此基础上提出一种基于神经网络的超声背散射零差K成像脂肪肝评价方法。方法:首先,利用蒙特卡罗仿真得到模拟的超声背散射包络信号样本,然后提取特征参数并输入到训练好的神经网络模型中,即可得到零差K模型参数的估计结果。其次,利用滑动窗口法估计窗口内局部背散射包络信号的零差K模型参数值,对零差K模型参数估计值矩阵进行扫描变换、颜色映射和感兴趣区域设置,将感兴趣区域内的参数图像叠加显示到B超图像中,实现超声背散射零差K成像。最后,通过计算机仿真实验验证基于神经网络的超声背散射零差K模型参数估计方法的估算精度,通过临床实验验证基于神经网络的超声背散射零差K成像评价脂肪肝的性能。结果:计算机仿真实验结果表明,基于神经网络的超声背散射零差K模型参数估计方法估计零差K模型lg α参数、k参数的相对均方根误差分别为0.505和0.408,相比传统估计方法整体上提高了估算精度。临床实验结果表明,基于神经网络的超声背散射零差K模型参数αNN、kNN诊断脂肪肝≥S1、≥S2、≥S3的AUC值分别为0.77、0.84、0.87和0.77、0.84、0.84,相比基于传统超声背散射零差K成像提高了脂肪肝评估性能。结论:提出的基于神经网络的超声背散射零差K成像脂肪肝评价方法能够较好地评估脂肪肝,可为定量超声评价脂肪肝提供一种新手段。Objective To propose two neural network estimator-based methods respectively for estimating ultrasound backscatter homodyned-K model parameters and evaluating hepatic steatosis with ultrasound backscatter homodyned-K imaging to solve the problem of high complexity of the traditional estimation method of ultrasound backscatter homodyned-K model parameters. Methods Firstly, simulated ultrasonic backscatter envelope signal samples were obtained using Monte Carlo simulation, and then the feature parameters were extracted and input to the trained neural network model to gain the estimation results of homodyned-K model parameters. Secondly, the homodyned-K model parameters for the backscatter envelope signals within the windows were estimated by the sliding window method, and the matrix of the estimated homodyned-K model parameters underwent scaning transformation, color mapping and setting of region of interest, and the parameter images in the region of interest were superimposed onto the B-ultrasound images to realize ultrasound backscatter homodyned-K imaging. Finally, the estimation accuracy of the neural network-based parameter estimation method for ultrasound backscatter homodyned-K model was verified by computer simulation experiments, and the performance of the neural network-based ultrasound backscatter homodyned-K imaging for hepatitis steatosis assessment was validated by clinical trials. Results The results of computer simulation experiments showed that the relative root-mean-square errors of estimating ultrasound backscatter homodyned-K model parameters lg α and k by the neural network-based method were0.505 and 0.408, respectively, and the estimation accuracy was enhanced significantly when compared with the traditional estimation method. The results of clinical trials indicated that the AUC values of the neural network-based ultrasound backscatter homodyned-K model parameters αNNand kNNfor detecting steatosis ≥S1, ≥S2, and ≥S3 were 0.77, 0.84, 0.87and 0.77, 0.84, 0.84, respectively, and the perfo

关 键 词：神经网络 超声背散射 定量超声成像 肝脏脂肪变性 零差K模型 脂肪肝 

分 类 号：R318[医药卫生—生物医学工程] R445[医药卫生—基础医学]