基于叠堆式LSTM网络的肺部电阻抗成像算法研究OA
Research on stacked LSTM network-based algorithm of pulmonary electrical impedance tomography
目的:为了提升肺部电阻抗成像(electrical impedance tomography,EIT)图像质量,提出基于叠堆式长短期记忆(long short-term memory,LSTM)网络的EIT算法.方法:首先,利用LSTM网络或双向长短期记忆(bidirectional long short-term memory,BiLSTM)网络初次重建EIT图像,初步重建电导率分布;其次,将重建电导率分布作为另一LSTM网络或BiLSTM网络的输入序列,再次重构新的电导率分布和EIT图像,并得到L-LSTM网络、L-BiLSTM网络、BI-LSTM网络和BI-BiLSTM网络4种叠堆式LSTM网络;最后,基于有限元分析构建圆截面模型和肺截面模型评估叠堆式LSTM网络的图像重建性能,并比较初次重建的EIT图像和再次重建的EIT图像的伪影与质量.结果:相较于LSTM网络,L-LSTM网络和L-BiLSTM网络的相关系数最大增幅分别为29.7%和29.5%,EIT图像中伪影大幅减少;相较于BiLSTM网络,BI-LSTM网络和BI-BiLSTM网络的相关系数最大增幅分别为11.7%和13.4%,EIT图像中伪影较少.L-LSTM网络和L-BiLSTM网络的整体性能优于BI-LSTM网络和BI-BiLSTM网络.结论:提出的基于叠堆式LSTM网络的EIT算法可获得伪影更少、图像质量更高的EIT图像,为LSTM网络及BiLSTM网络在EIT技术中的应用提供了新的优化思路.
Objective To propose an electrical impedance tomography(EIT)algorithm based on the stacked long short-term memory(LSTM)network so as to improve pulmonary EIT image quality.Methods Firstly,EIT images and conductivity distribution were reconstructed initially with a LSTM or bidirectional LSTM(BiLSTM)network;secondly,the reconstructed conductivity distribution was fed as an input sequence of another LSTM or BiLSTM network to further refine the conductivity distribution and produce updated EIT images,thereby yielding four types of stacked LSTM networks,namely L‑LSTM,L‑BiLSTM,BI‑LSTM and BI‑BiLSTM.Finally,finite‑element analysis was used to construct circular and lung cross‑section models to assess the imaging performance of the stacked LSTM networks,and to compare the initially and subsequently refined EIT images in terms of the artifact and quality.Results The L-LSTM network and L-BiLSTM network had the maximum increase of correlation coefficients being 29.7%and 29.5%respectively and the artifacts in the EIT images reduced greatly when compared with the LSTM network;the BI-LSTM network and BI-BiLSTM network had the maximum increase being 11.7%and 13.4%respectively and the artifacts decreased when compared with the BiLSTM network.The L-LSTM and L-BiLSTM networks behaved better generally than the BI-LSTM and BI-BiLSTM networks.Conclusion The proposed algorithm based on the stacked LSTM network contributes to obtaining EIT images with few artifacts and high quality,which provides a new optimization idea for the LSTM and BiLSTM networks applied to EIT technologies.[Chinese Medical Equipment Journal,2026,47(9):1-9]
宋振忠;李建平;李易滴;张俊;杨庆虎;文汉云;周星星;代芳
贵州中医药大学时珍学院,贵阳 550200浙江师范大学,浙江 金华 321004贵州中医药大学时珍学院,贵阳 550200贵州中医药大学时珍学院,贵阳 550200贵州中医药大学时珍学院,贵阳 550200贵州中医药大学时珍学院,贵阳 550200贵州中医药大学时珍学院,贵阳 550200贵州中医药大学时珍学院,贵阳 550200
医药卫生
肺部电阻抗成像长短期记忆网络双向长短期记忆网络重建算法叠堆式LSTM网络
pulmonary electrical impedance tomographylong short-term memory networkbidirectional long short-term memory networkreconstruction algorithmstacked LSTM network
《医疗卫生装备》 2026 (9)
1-9,9
贵州中医药大学时珍学院教学改革工程项目(SZXY2024005)
评论