Detection of Suspicious Cardiotocographic Recordings by Means of a Machine Learning Classifier

Area of Science:

• Obstetrics and Gynecology
• Medical Informatics
• Machine Learning in Healthcare

Background:

• Cardiotocography (CTG) is crucial for prenatal fetal surveillance, yet its diagnostic accuracy for uncertain traces remains a challenge.
• Computerized analysis systems offer potential for more objective and accurate CTG evaluations.
• Distinguishing suspicious from normal CTG traces is a persistent difficulty for clinicians.

Purpose of the Study:

• To address the challenge of classifying suspicious cardiotocography (CTG) recordings using a machine learning approach.
• To develop and evaluate a machine-based labeling and binary classification system for CTG traces.
• To differentiate between suspicious and normal CTG recordings for enhanced fetal monitoring.

Main Methods:

• A machine learning approach was employed, utilizing a Support Vector Machine (SVM) classifier.
• A binary classification task was performed to distinguish between suspicious and normal CTG traces.
• Machine-based labeling was proposed and implemented for the classification process.

Main Results:

• The Support Vector Machine (SVM) classifier achieved high performance metrics.
• Classification accuracy reached 92%, sensitivity was 92%, and specificity was 90%.
• Results were validated against unbalanced datasets and existing literature.

Conclusions:

• The application of Support Vector Machines (SVM) shows significant potential for CTG classification.
• Effective feature selection and dataset balancing are critical for optimizing classifier performance.
• This machine learning approach offers a promising avenue for improving the accuracy of fetal surveillance through CTG analysis.