Fully Automatic Classification of Cardiotocographic Signals with 1D-CNN and Bi-directional GRU

Area of Science:

• Obstetrics and Gynecology
• Medical Signal Processing
• Artificial Intelligence in Healthcare

Background:

• Prenatal fetal monitoring is crucial for assessing fetal growth and health before delivery.
• Accurate classification of fetal abnormalities aids in early intervention, preventing fetal heart hypoxia and mortality.
• Fetal heart rate and uterine contraction signals are key indicators of fetal well-being.

Purpose of the Study:

• To develop and evaluate a novel method for classifying fetal abnormalities using signal processing and deep learning.
• To enhance the accuracy and reliability of prenatal fetal monitoring systems.
• To provide physicians with improved decision support for managing high-risk pregnancies.

Main Methods:

• Data preprocessing and enhancement using Hermite interpolation for abnormal classification.
• Implementation of a hybrid 1D-CNN (Convolutional Neural Network) and GRU (Gated Recurrent Unit) model.
• Extraction of abstract features from fetal heart rate and uterine contraction signals.

Main Results:

• The proposed hybrid model achieved high performance metrics: 96% accuracy, 95% sensitivity, and 96% specificity.
• Hermite interpolation effectively enhanced the classification of abnormal fetal samples.
• The model demonstrated robust feature extraction capabilities for complex physiological signals.

Conclusions:

• The developed method offers a stable, efficient, and convenient approach for prenatal fetal health diagnosis.
• This technique can significantly aid physicians in making timely and informed clinical decisions.
• The findings highlight the potential of hybrid deep learning models in improving prenatal care outcomes.