Optimizing Clinical Diabetes Diagnosis through Generative Adversarial Networks: Evaluation and Validation.

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Data Science

Background:

• Type 2 Diabetes (T2D) prevalence is increasing globally, posing a significant healthcare challenge.
• Current T2D diagnostic methods can be invasive and labor-intensive.
• Limited patient datasets hinder the application of advanced AI and data science for T2D diagnosis.

Purpose of the Study:

• To investigate the effectiveness of Generative Adversarial Networks (GANs) in augmenting Type 2 Diabetes (T2D) patient data.
• To improve the diagnostic accuracy of machine learning models for T2D using synthetic data.
• To address data scarcity issues in clinical AI applications for T2D.

Main Methods:

• Utilized a dataset of 1019 Mexican nationals (499 non-diabetic, 520 T2D cases).
• Applied Generative Adversarial Networks (GANs) to create synthetic patient data.
• Trained a Random Forest (RF) classification model using augmented and original data.

Main Results:

• GAN-based data augmentation led to a significant improvement in the diagnostic accuracy of the Random Forest model.
• Synthetic data effectively supplemented the limited real-world patient data.
• Validated the utility of GANs for enhancing machine learning model performance in a clinical setting.

Conclusions:

• Generative Adversarial Networks (GANs) show promise for overcoming data limitations in AI-driven T2D diagnosis.
• Data augmentation using GANs can enhance the robustness and reliability of machine learning tools for T2D.
• This approach offers a potential pathway to more timely and effective T2D diagnosis and patient management.