Clinical utility assessment framework for machine learning-based fetal health classification in cardiotocography: an...

Area of Science:

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

Background:

• Artificial intelligence (AI) systems are increasingly explored for fetal health classification.
• Cardiotocography (CTG) data is crucial for monitoring fetal well-being.
• Evaluating the clinical utility and implementation of AI in obstetrics is essential.

Purpose of the Study:

• To assess the clinical utility and implementation of AI-based fetal health classification systems.
• To analyze obstetric physicians' perspectives on AI in fetal health monitoring.
• To evaluate machine learning models using the Kaggle Fetal Health Classification dataset.

Main Methods:

• Analysis of the Kaggle Fetal Health Classification dataset (n=2,126) with 21 CTG parameters.
• Evaluation of five machine learning algorithms: logistic regression, random forest, gradient boosting, SVM, and decision tree.
• Development of a clinical utility assessment framework based on expert opinion, focusing on interpretability, workflow integration, and safety.

Main Results:

• Gradient boosting achieved the highest accuracy (89.67%), followed by random forest (88.50%).
• Abnormal short-term variability and percentage of time with abnormal long-term variability were key predictive features.
• A significant false negative rate (35.3%) for pathological cases highlights safety concerns, with contraction parameters contributing minimally.

Conclusions:

• AI-based fetal health classification systems demonstrate potential but require rigorous validation.
• The current false negative rates preclude independent system operation, emphasizing the need for physician oversight.
• External validation with multicenter clinical data and prospective studies is crucial before widespread clinical implementation.