Prediction of gestational diabetes using deep learning and Bayesian optimization and traditional machine learning...

Area of Science:

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

Background:

• Gestational diabetes (GD) diagnosis relies on oral glucose tolerance tests (OGTT), which can be resource-intensive and lead to unnecessary procedures for low-risk pregnancies.
• Developing accurate, non-invasive diagnostic tools is crucial for efficient prenatal care and reducing healthcare costs.

Purpose of the Study:

• To develop and validate a deep learning-based clinical decision support system for identifying pregnant women at risk of gestational diabetes (GD).
• To reduce the application of unnecessary oral glucose tolerance tests (OGTT) in pregnant women not belonging to the GD risk group.

Main Methods:

• A prospective study involving 489 patients between 2019-2021.
• Development of a clinical decision support system using deep learning algorithms, specifically RNN-LSTM, combined with Bayesian optimization.
• Model training and validation on the collected patient dataset.

Main Results:

• A novel decision support model utilizing RNN-LSTM with Bayesian optimization was successfully developed.
• The model achieved 95% sensitivity and 99% specificity for identifying patients in the GD risk group.
• The model demonstrated high diagnostic accuracy with an Area Under the Curve (AUC) of 98% (95% CI: 0.95-1.00, p < 0.001).

Conclusions:

• The developed clinical diagnosis system effectively assists physicians in identifying gestational diabetes risk.
• The system has the potential to significantly save costs and time while minimizing adverse effects by preventing unnecessary OGTTs for low-risk individuals.