Research on CTG Feature Extraction Algorithm Based on Deep Learning

Cardiotocograph (CTG) is a commonly used method by obstetricians and gynecologists for monitoring pregnant women during the perinatal period. Doctors can analyze the relevant characteristics of fetal heart rate (FHR) with CTG under the guidance of CTG guidelines, which is crucial for the subsequent evaluation of fetal status and directly affects the choice of delivery method. However, clinical doctors often have disagreements in interpreting signal characteristics. To address these challenges, this study adopts a computer-aided approach leveraging advancements in deep learning. An innovative feature analysis of CTG signals is proposed using the Ensemble Attention Dense U-Net (EADU-Net) network given the popularity of high-performance deep learning in recent years. For the ongoing controversy surrounding the interpretation of CTG signals, the CBAM attention mechanism and the Dense-block module's Ensemble Attention Dense U-Net (EADU-Net) network for automatic feature extraction of FHR signals from CTG signals was introduced. 66 data provided by the French Lille database for training and validation were used. The trained model was tested using 331 data from the Overseas Chinese Hospital database to obtain five relevant evaluation indicators. The performance of the algorithm was quantitatively compared with the traditional algorithm (WMFB algorithm) that currently has good performance. According to the test data, the EDAU-Net algorithm proposed in this paper improved the F measure (Acc. F-measure) by 17%, the Dec. F-measure (Dec. F-measure) by 19.5%, the Root mean-square difference of baselines (BL. RMSD) by 0.17bpm, the Synthetic inconsistency coefficient (SI) by 16.85%, and the Morphological analysis discordance index (MADI) by 0.55%. These findings indicate that compared with traditional signal processing-based algorithms, the CTG signal feature automatic extraction algorithm has the smallest difference in discrimination with clinical obstetrics and gynecology experts.