Real-time Classification of Fetal Status Based on Deep Learning and Cardiotocography Data

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Fetal Monitoring

Background:

• Cardiotocography (CTG) is crucial for assessing fetal well-being.
• Real-time analysis of CTG data can improve timely clinical decisions.
• Current methods may lack accessibility or real-time processing capabilities.

Purpose of the Study:

• To develop and evaluate a deep learning model for simultaneous real-time fetal status classification.
• To compare the performance of 1D CNN on a mobile application with 2D CNN on a computer server.
• To assess the efficiency and accuracy of convolutional neural networks (CNNs) in analyzing cardiotocography (CTG) data.

Main Methods:

• Utilized 141,001 CTG traces from 5249 (or 4833) patients.
• Trained 1-dimensional Resnet CNN for mobile application and 2-dimensional Resnet CNN for computer server.
• Fetal heart rate and uterine contractions were used as predictors for classification into Normal/Abnormal or Normal/Middle/Abnormal fetal status.

Main Results:

• The 1D CNN achieved 94.9% sensitivity, 91.2% specificity, and 93.0% harmonic mean.
• The 2D CNN achieved 98.0% sensitivity, 99.5% specificity, and 98.7% harmonic mean.
• Average inference time was 6.51 microseconds per 1000 images, demonstrating high efficiency.

Conclusions:

• Deep learning models, specifically CNNs, offer an efficient and accurate method for real-time fetal status classification.
• The developed system provides simultaneous real-time analysis for both pregnant women via mobile app and data experts via computer server.
• This technology has the potential to significantly enhance prenatal care and monitoring accessibility.