Related Experiment Video
Updated: May 11, 2025
06:18
Author Spotlight: Segmentation and VR for Advanced Neurovascular Interventions
Published on: April 5, 2024
915
Generating Artificial Patients With Reliable Clinical Characteristics Using a Geometry-Based Variational Autoencoder:
Fabrice Ferré1, Stéphanie Allassonnière2, Clément Chadebec2
1Department of Anesthesia, Intensive Care and Perioperative Medicine, Purpan University Hospital, Toulouse, France.
Journal of Medical Internet Research
|April 17, 2025
Summary
Researchers generated artificial patients using variational autoencoders (VAE) on tabular data. This technology creates reliable synthetic patient data for healthcare, ensuring patient confidentiality and enabling in silico trials.
Area of Science:
- Artificial intelligence in healthcare
- Machine learning for synthetic data generation
- Computational biology and medicine
Background:
- Artificial patient technology offers transformative potential in healthcare for accelerating diagnosis and treatment.
- Deep learning, specifically variational autoencoders (VAE), is a key method for generating artificial health data.
Purpose of the Study:
- To assess the feasibility of creating artificial patients with reliable clinical attributes using a geometry-based VAE.
- To apply VAEs to high-dimension, low-sample-size tabular data for the first time.
Main Methods:
- Extracted clinical tabular data from 521 real patients for anesthesia preparation.
- Implemented a three-stage approach: model training/data generation, consistency/confidentiality assessment, and plausibility validation.
- Generated up to 10,000 artificial patients.
Main Results:
- Demonstrated VAE feasibility on tabular data, generating large cohorts with over 94% fidelity.
- Ensured patient confidentiality, with artificial patients unmatchable to real patients (similarity scores >99%).
Conclusions:
- Proof-of-concept study successfully augmented real tabular data to generate artificial patients.
- Results support the potential for in silico trials on large artificial patient cohorts, overcoming in vivo trial limitations.
- Further research is needed to incorporate longitudinal data for patient trajectory mapping.
