Generating Synthetic Medical Data Using GANs
Author Information
Author(s): Nasimov Rashid, Nasimova Nigorakhon, Mirzakhalilov Sanjar, Tokdemir Gul, Rizwan Mohammad, Abdusalomov Akmalbek, Cho Young-Im
Primary Institution: Tashkent State University of Economics
Hypothesis
Can statistical data be used to generate synthetic medical tabular data using a modified GAN architecture?
Conclusion
The proposed method successfully generates synthetic medical data that maintains statistical properties similar to real datasets, achieving good fidelity and utility scores.
Supporting Evidence
- The proposed method achieved 'Good' similarity and 'Excellent' utility scores.
- Statistical tests showed that the generated data maintained similar means to the real dataset.
- The method allows for synthetic data generation without direct access to real patient data.
Takeaway
This study shows how we can create fake medical data that looks real by using numbers and statistics instead of actual patient records, helping to keep people's information private.
Methodology
The study used a modified Generative Adversarial Network (GAN) architecture with a custom loss function to generate synthetic tabular data from statistical inputs.
Potential Biases
The method relies on the quality of the statistical data used, which may introduce biases if the statistical data is not representative.
Limitations
The generated data may not fully replace real databases and may not capture all variability of real-world medical data.
Participant Demographics
The study utilized the Pima Diabetes Database, which includes data from Pima Indians in Arizona, USA.
Statistical Information
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website