LaborEase: An Artificial Intelligence-Based Wearable Electrohysterography Device for Preterm Pregnancy Detection
Muhammad Omar Cheema, Zia Mohy Ud Din, Abdullah Al Aishan, Jahan Zeb Gul
Abstract
In this research, a portable electrohysterography device is developed for electrohysterogram recording and early detection of preterm pregnancies using machine learning techniques to classify the acquired recordings as preterm or term. The data collected in this study is combined with ‘TPEHGDB’ open-source dataset. With the aid of SMOTE ENN, Multilayer Perceptron achieved the highest 10-fold Cross-Validation Accuracy of 0.90, Test Accuracy, Recall, Precision, and F1 Score of 0.96, Sensitivity of 0.97, Specificity and ROC of 0.98, Cohen’s Kappa and MCC of 0.91 when various features are used for classification and the Random Forest classifier achieved the highest 10-fold Cross-Validation Accuracy of 0.95, Test Accuracy, Recall, Precision, F1 Score, Sensitivity, and Specificity of 0.94, ROC of 0.98, Cohen’s Kappa andMCCof 0.87 when only optimized features and channels are used for classification. The proposed device has a signal-to-noise ratio of 20.44 dB ensuring the quality of the acquired signals. The developed system can be used for early, noninvasive, reliable, and cost-effective detection of preterm pregnancies, enhancing maternal and neonatal health monitoring and saving millions of lives with cost-effective and proactive interventions.