Evaluation of convolutional neural network for recognizing uterine contractions with electrohysterogram

Area of Science:

• Biomedical Engineering
• Signal Processing
• Machine Learning in Healthcare

Background:

• Uterine contraction (UC) activity monitoring is crucial for assessing labor and delivery progress.
• Electrohysterograms (EHGs) offer a non-invasive method for monitoring UC, distinguishing between efficient and inefficient contractions.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) model for identifying uterine contractions (UCs) within electrohysterogram (EHG) signals.
• To assess the generalizability of the developed CNN model on an independent clinical dataset.

Main Methods:

• A CNN model was developed using a 16-electrode EHG database (DB1) comprising 14,000 segments (7,000 UC, 7,000 non-UC).
• The model's performance was evaluated using five-fold cross-validation, assessing sensitivity (SE), specificity (SP), accuracy (ACC), and area under the ROC curve (AUC).
• The validated CNN model was tested on a separate clinical database (DB2) with 308 EHG segments (154 UC, 154 non-UC).

Main Results:

• The CNN model achieved high performance metrics on DB1: average SE of 0.87, SP of 0.98, ACC of 0.93, and AUC of 0.92.
• On the independent DB2, the model demonstrated robust generalizability with average SE of 0.88, SP of 0.97, ACC of 0.93, and AUC of 0.87.
• The results indicate the CNN's effectiveness in accurately identifying UCs from EHG signals.

Conclusions:

• Convolutional neural networks (CNNs) can effectively identify uterine contractions (UCs) using electrohysterogram (EHG) signals.
• The developed CNN model demonstrates strong performance and generalizability for UC detection.
• This approach holds potential as a valuable tool for monitoring maternal and fetal well-being.