Augmenting Blood Test and Periodontal Examination Data with Generative Adversarial Networks for Enhanced Dementia...

Area of Science:

• Medical Informatics
• Artificial Intelligence in Healthcare
• Neurology

Background:

• Blood tests and Mini-Mental State Examination (MMSE) scores are vital for cognitive function prediction.
• Medical data acquisition faces challenges like high costs, small sample sizes, and missing data.
• Periodontal examinations offer a cost-effective method for mass screening.

Purpose of the Study:

• To evaluate the effectiveness of data augmentation using generative adversarial networks (GANs) for dementia risk prediction.
• To compare the accuracy of deep neural networks (DNNs) trained on real versus GAN-synthesised medical data.
• To assess the robustness of models using synthesised data against missing variables and data imbalances.

Main Methods:

• Utilised deep neural networks (DNNs) with four hidden layers for dementia risk prediction.
• Employed generative adversarial networks (GANs) to synthesise data from blood tests and periodontal examinations.
• Compared prediction accuracy (Mean Absolute Error - MAE) between models trained on real and GAN-synthesised data from 108 participants.

Main Results:

• GAN-synthesised DNNs achieved a lower Mean Absolute Error (MAE) of 1.91 ± 0.30 compared to real data (2.04 ± 0.37).
• Synthesised data demonstrated improved robustness against missing variables, including age.
• The augmentation approach effectively managed data imbalances in the dataset.

Conclusions:

• Data augmentation with GANs significantly enhances dementia risk prediction accuracy.
• Synthesised medical data improves model performance and robustness, addressing common data limitations.
• This augmentation strategy shows considerable potential for refining dementia risk prediction models in resource-constrained settings.