基于人工神经网络的零差K参数成像监测微波热消融凝固区研究

Microwave ablation coagulation area monitoring by homodyned K imaging based on artificial neural network

目的:研究基于人工神经网络(artificial neural network,ANN)的零差K参数成像监测微波热消融凝固区的可行性。方法:在消融功率和持续时间分别为60 W 3 min、70 W 2 min、80 W 1 min的条件下,各采集16例经过微波热消融的离体猪肝的超声背散射信号,分别利用ANN估计器和XU估计器在滑动窗口内估算背散射信号零差K分布模型的α、k参数,之后与多项式拟合技术相结合监测微波热消融产生的凝固区,并比较基于2种估计器获得的零差K参数成像监测微波热消融凝固区的效果。结果:在不同消融功率、持续时间条件下,α参数成像对微波热消融凝固区的监测性能优于k参数成像;2种估计器均在80 W 1 min时获得了最好的监测效果;基于ANN估计器和XU估计器获得的基于α参数的多项式拟合成像的AUC值、Dice系数、Jaccard系数差异均无统计学意义(P均>0.05)。结论:基于ANN的零差K参数成像可用于监测微波热消融凝固区。

Objective To investigate the feasibility of homodyned K parametric imaging based on an artificial neural network(ANN)for monitoring the coagulation zone during microwave ablation.Methods Ultrasound backscatter signals were collected from 16 porcine livers ex vivo subjected to microwave ablation under the conditions of ablation power and duration of 60 W for 3 min,70 W for 2 min and 80 W for 1 min.Theαand k parameters of the homodyned K distribution model of backscatter signals were estimated within a sliding window using ANN estimator and XU estimator,and then were combined with polynomial approximation technique to monitor the coagulation zone,the two estimators were compared in terms of the monitoring effect of the microwave ablation coagulation zone.Resultsαparameter imaging behaved better than k parameter imaging under different conditions of ablation power and duration;both of the two estimators achieved the best monitoring results with 80 W 1 min microwave ablation.There were no statistically significant differences between the two estimators in terms of AUC value,Dice coefficient and Jaccard coefficient of the polynomial approximation image based onαparameter(all P>0.05).Conclusion ANN-based homodyned K parametric imaging can be used to monitor the coagulation zone during microwave ablation.[Chinese Medical Equipment Journal,2023,44(5):1-9]