Enhancing healthcare information security through VAE-driven anomaly detection in EHR access patterns

Area of Science:

• Computer Science
• Healthcare Informatics
• Cybersecurity

Background:

• Digitalization of healthcare increases risks of HIPAA breaches and unauthorized access to Electronic Health Records (EHRs).
• Effective security models are challenged by limited access to real-world healthcare data due to privacy constraints.

Purpose of the Study:

• To propose a privacy-preserving anomaly detection model for identifying suspicious access patterns in EHR systems.
• To address the challenge of limited real-world data by utilizing a synthetically enriched EHR access log dataset.

Main Methods:

• Developed a Variational Autoencoder (VAE) based anomaly detection model.
• Generated a synthetic EHR access log dataset with realistic features (e.g., user roles, timestamps).
• Evaluated the VAE model against traditional machine learning algorithms like Isolation Forest and One-Class SVM.

Main Results:

• The VAE model demonstrated superior performance in detecting anomalies, achieving an F1-score of 0.93.
• Achieved lower false-positive rates and greater sensitivity to noisy data compared to traditional methods.
• The model effectively identified deviations in access patterns without accessing sensitive patient data.

Conclusions:

• Unsupervised deep generative modeling combined with synthetic data generation offers a novel, privacy-conserving approach to enhance medical information system cybersecurity.
• VAE-based anomaly detection is a promising solution for protecting healthcare infrastructure against evolving cyber risks.