基于超声射频信号的支持向量机双参量B线识别方法*

肺超声中的特殊征象B线对于临床诊断肺水肿等肺部疾病有重要意义，但诊断结果依赖于医生的主观判断，为了客观、自动地识别B线，提高诊断准确率，本文提出了一种基于超声回波射频信号的肺脏超声特殊征象B线识别方法。本文首先选取了射频信号的排列熵、信息熵、峰度、偏度、能量作为特征参数，利用独立样本t检验和单参数贝叶斯分类的方法检验超声射频数据中B线以及非B线所对应射频数据的各个参量的差异性以及各参数与B线识别的相关性。然后将不同的双参量组合输入非线性支持向量机（SVM）中进行分类，比较各个组合的分类效果。结果显示信息熵与排列熵参数组合基于射频信号的分类效果最好，分类灵敏度为90.521%，特异性为98.106%，准确率为96.328%，AUC等于0.95。在引入后处理算法后，B线识别效果有进一步提升，得到分类平均灵敏度为95.23%，平均特异性为97.22%，平均准确率为96.88%。研究结果表明基于射频数据的SVM双参量B线识别方法对辅助临床诊断具有重要价值，信息熵和排列熵的组合可以有效的对特殊征象B线进行高精度识别。

Support vector machine-based two-parameter B-line identification method using ultrasound radio frequency signal

Special artifact B-line of lung ultrasound has important significance for clinical diagnosis of lung lesions like pulmonary edema and so on. However, accurate diagnosis depends on subjective judgment of the doctor. In order to identify B line objectively and automatically, so as to improve the diagnostic accuracy, a B-line identification method for lung ultrasound using ultrasonic echo radio frequency (RF) signals was proposed. The t-test and Bayes test methods were used to test distinction and Identification correlation of several characteristic parameters including permutation entropy, information entropy, kurtosis, skewness and energy between B-Line and non-B-Line area. Different two-parameter combinations were input into nonlinear SVM for classification. The classification results of these combinations were compared. The results showed that the two-parameter combination of information entropy and permutation entropy parameters had the best classification effect based on RF signals, with a sensitivity of 90.521%, specificity of 98.106%, accuracy of 96.328% and AUC of 0.95. After the post-processing algorithm, the identification effect of B-line was further improved, and the average sensitivity, specificity and accuracy of classification were 95.23%, 97.22% and 96.88% respectively. The results showed that the SVM two-parameter B-line identification method based on RF data had important value in assisting clinical diagnosis, and the combination of information entropy and permutation entropy could identify the B-line with high precision.