文献阅读7

 08

Date:2022.12.24--09
Title: A deep feature fusion network for fetal state assessment Link: Frontiers | A deep feature fusion network for fetal state assessment (frontiersin.org) ​
Dataset	CTU-UHB: only FHR and JNU-STG（暨大私有）
Framework	​​编辑 ​​
Methodology	In this paper, we propose a novel deep feature fusion network (DFFN) for fetal state assessment. First, we extract spatial and temporal information from the fetal heart rate (FHR) signal using a multiscale CNN-BiLSTM network, increasing the features’ diversity. Second, the multiscale CNN-BiLSM network and frequently used features are integrated into the deep learning model.
Results	The proposed method achieves 61.97%, 73.82%, and 66.93% of sensitivity, specificity, and quality index, respectively, ​ ​
Contributions	The chief contributions are summarized as follows. 1) As far as we know, this work is the first to use a deep feature fusion network (DFFN) that combines a multiscale CNN-BiLSTM model with linear and nonlinear features to improve the classification performance. 2) The multiscale CNN-BiLSTM model simultaneously derives spatial features and temporal information from CTG signals to capture complex fetal vital signs. 3) We construct the JNU-CTG database and use it to validate the generalizability of the proposed method.   Compared to other researches, the present method has the best generalization performance.
Conclusion	This paper proposes a novel deep feature fusion network for diagnosing fetal acidosis from FHR signals. A multiscale CNNBiLSTM hybrid network is developed to extract the signal’s temporal and spatial features adequately. In order to account for clinical physiological parameters and assessment accuracy, a feature fusion network is used to splice the multiscale CNNBiLSM features, as well as the currently popular linear and nonlinear features.    The experimental results on two databases show that DFFN achieves better performance than previous works. The accuracy of fetal state classification as well as the generalization of DFFN are improved by merging the FHR features from multiscale layers with the extra features.
Notes	​​ ​​