Cardiac aging synthesis from cross-sectional data with conditional generative adversarial networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Health

Background:

• Age is a primary risk factor for cardiovascular disease.
• Understanding cardiac aging is crucial for potential interventions.
• Longitudinal medical imaging data is costly and difficult to obtain for deep learning.

Purpose of the Study:

• To develop a method for synthesizing older and younger heart scans using only cross-sectional data.
• To model cardiac aging changes realistically without longitudinal datasets.
• To demonstrate the utility of synthetic data for correcting age bias in medical imaging datasets.

Main Methods:

• A conditional generative adversarial network (cGAN) was employed to synthesize cardiac images.
• The model was trained on over 14,000 heart scans from the UK Biobank.
• Image quality, predicted age accuracy, and bias correction capabilities were evaluated.

Main Results:

• The cGAN successfully generated realistic older and younger heart scans.
• Synthesized modifications primarily affected the interventricular septum and aorta, aligning with known cardiac aging patterns.
• The generated synthetic data effectively counter-balanced age bias in a dataset.

Conclusions:

• The proposed cGAN approach can effectively model cardiac aging using cross-sectional data.
• This method offers a cost-effective way to generate diverse cardiac imaging data.
• Synthetic data generated by this model can be valuable for training AI models and mitigating dataset bias.