Generation of Individualized Synthetic Data for Augmentation of the Type 1 Diabetes Data Sets Using Deep Learning...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Diabetes Technology

Background:

• Continuous glucose monitoring (CGM) is crucial for managing type 1 diabetes mellitus (T1DM).
• Limited availability of diverse patient data can hinder the performance of machine learning (ML) models for predicting glycemic events.
• Generating synthetic patient data offers a potential solution to augment real-world datasets.

Purpose of the Study:

• To develop a generative adversarial network (GAN) methodology for synthesizing CGM data.
• To evaluate the capability of GANs to replicate individual patient characteristics and statistical distributions.
• To assess the impact of synthetic CGM data on improving ML model performance for predicting nocturnal hypoglycemia in T1DM.

Main Methods:

• A GAN architecture was employed to generate synthetic CGM data from two distinct T1DM patient cohorts.
• The fidelity of the synthetic data was assessed by comparing its statistical distributions to the original data.
• Multiple ML prediction models were trained and compared using both original and augmented datasets to predict nocturnal hypoglycemia.

Main Results:

• The GAN methodology successfully replicated intrinsic patient characteristics and statistical properties of the original CGM data.
• The use of synthetic data for model training led to improved performance in predicting nocturnal hypoglycemic events.
• Comparative analysis demonstrated the enhanced predictive power of ML models trained on augmented datasets.

Conclusions:

• Generative adversarial networks provide a viable method for creating realistic synthetic CGM data.
• Synthetic CGM data augmentation can significantly enhance the performance of ML models for T1DM glycemic event prediction.
• This approach sets a precedent for utilizing generative techniques in training ML models for diabetes management.