Anomaly Detection for Medical Data Synthesis Evaluation: A Case Study on Opioid Misuse

Poster Number: P198
Presentation Time: 05:00 PM - 06:30 PM

Abstract Keywords: Governance of Artificial Intelligence, Machine Learning, Delivering Health Information and Knowledge to the Public, Population Health, Data Sharing, Privacy and Security, Deep Learning
Primary Track: Applications
Programmatic Theme: Public Health Informatics

Synthetic medical data has been shown to bridge the gap between medical data security, privacy, and data sharing. Over the past years, a massive of data synthesis approaches have been developed driven by rapid developments of generative AI, especially with such as generative adversarial networks (GANs) models. However, little attention has been paid to the quality evaluation of these newly developed models. This paper takes advantage of anomaly detection methods, which do not require prior knowledge or groundtruth information from the dataset, to evaluate the quality of synthetic data generation. The proposed framework enables unsupervised evaluation of anomaly transfer efficiency between real and synthetic data. Using opioid misuse data as a case study, we generated a corresponding synthetic dataset and conducted a comprehensive investigation into the performance of anomaly detection transfer between the real and synthetic datasets.

Speaker(s):
Yili Zhang, PhD
Georgetown University

Author(s):
Yili Zhang, PhD - Georgetown University; Bai Xue, PhD - University of Maryland, Baltimore County; Jia Li Dong, MS - Georgetown University; Yanbao Xiong, MS - Medstar Health; Samir Gupta, PhD - Georgetown University; Maarten Van Segbroeck, PhD - Gretel.ai; Nawar Shara, PhD; Peter McGarvey, PhD - Georgetown University Medical Center;