Automated interpretation of cardiotocography using deep learning in a nationwide multicenter study.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Obstetrics and Gynecology

Background:

• Timely detection of abnormal cardiotocography (CTG) is vital for fetal prognosis during labor.
• Existing deep learning models for CTG interpretation often use limited datasets or focus on non-clinical outcomes.
• Clinical applicability of deep learning in CTG analysis remains a challenge.

Purpose of the Study:

• To develop and validate a clinically applicable deep learning model for automated CTG interpretation.
• To address limitations of previous studies by using a large-scale, nationwide dataset with expert annotations.
• To classify normal and abnormal CTG segments for improved fetal monitoring.

Main Methods:

• Utilized a large-scale dataset of 22,522 deliveries from 14 hospitals, with CTG recordings up to 75 minutes.
• Segmented CTG signals into 5-minute intervals, creating a dataset of 519,800 person-minutes.
• Trained and validated a deep learning model for binary classification of normal versus abnormal CTGs.

Main Results:

• The deep learning model achieved an AUC of 0.880 and PRC of 0.625 in internal tests.
• External validation across three datasets yielded AUCs ranging from 0.862 to 0.895 and PRCs from 0.553 to 0.615.
• Demonstrated robust performance in classifying CTG segments, indicating potential for automated interpretation.

Conclusions:

• Deep learning models show significant potential for automated interpretation of cardiotocography signals.
• The developed model, trained on a large, diverse dataset, exhibits strong performance in identifying abnormal CTGs.
• Future prospective studies are planned to further evaluate the clinical applicability of this automated CTG analysis tool.