New AI Model Protects Patient Privacy in ECG Data Without Compromising Heart Disease Detection

Advertisement

Privacy-Preserving Variational Autoencoder (PP-VAE) reduces the amount of personal information that AI can extract from ECG recordings while preserving their clinical value for detecting heart conditions.

Researchers at the University of Kansas have developed an AI model that helps protect patient privacy by limiting the exposure of personal information hidden within ECG data while maintaining accurate predictions for heart disease and mortality risk.

Called Privacy-Preserving Variational Autoencoder (PP-VAE), the technology reduces the amount of personal information that artificial intelligence can extract from ECG recordings while preserving their clinical value for detecting heart conditions and predicting health outcomes.

"We believe there are two key reasons people should use it," Fairuz Shadmani Shishir, a doctoral student in electrical engineering & computer science at KU who led the study, said. "First, the model is designed to generalize across patients in the United States. Second, we plan to make the model publicly available so that anyone can use it.”

“Making the model publicly available follows common practices in the AI field. Institutions will be able to use our model and potentially build their own versions trained on their own datasets. Our goal is to release the model publicly in the future," he added.

The research team designed the model to predict clinically significant outcomes, including left ventricular ejection fraction (LVEF), an important measure of heart function that is associated with heart abnormalities and early mortality risk.

The researchers evaluated the model using independent convolutional neural networks and found that it reduced the identifiability of soft biometric traits while maintaining reliable predictions for conditions such as left ventricular hypertrophy and five-year mortality.

The team also focused on reducing bias by including balanced representation across sex and racial groups during model development.

While the AI model was trained using data from the University of Kansas Medical Center and validated on public datasets, the researchers plan to expand future training using data from different regions to further improve fairness and generalizability.

Looking ahead, the researchers intend to make the model publicly available so healthcare institutions can adopt or further develop it using their own datasets, potentially enabling safer AI innovation while strengthening patient privacy protections across digital healthcare ecosystems.

Stay tuned for more such updates on Digital Health News