Related Experiment Video
Updated: Jan 9, 2026
07:15
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
7.3K
Utilizing synthetic data for privacy-preserving AI modeling in radiomics: a case study
Summary
Synthetic radiomics data can protect patient privacy in AI cancer classification. The Bayesian Gaussian Mixture Model approach generated high-fidelity synthetic data, enabling accurate prostate cancer aggressiveness prediction while preserving privacy.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Data Privacy and Security
- Oncology
Background:
- Healthcare AI models, especially in radiomics, face privacy challenges due to sensitive patient data.
- Radiomics data can contain unique identifiers, posing re-identification risks.
- Protecting patient privacy is paramount for ethical AI deployment in healthcare.
Purpose of the Study:
- To explore synthetic radiomics data as a privacy-preserving strategy for AI-driven prostate cancer aggressiveness classification.
- To evaluate the data fidelity and model performance of synthetic radiomics data generated by different methods.
- To assess the balance between data utility and privacy preservation in AI radiomics.
Main Methods:
- Radiomics features were extracted from Multiparametric MRI (mpMRI) data of 4,588 retrospective and 1,369 prospective prostate cancer cases across 12 EU centers.
- Three advanced generative models were investigated: Bayesian Gaussian Mixture Models with optimal components estimation (BGMMOCE), Conditional Tabular Generative Adversarial Network (CTGAN), and Tabular Variational AutoEncoder (TVAE).
- Data fidelity was quantified using Jensen-Shannon divergence (JSD) and Hellinger distance (HD); model performance was evaluated using a Random Forest (RF) classifier.
Main Results:
- The BGMMOCE model demonstrated superior data fidelity compared to CTGAN and TVAE, achieving a JSD of 0.08 and HD of 0.23.
- A Random Forest classifier trained on BGMMOCE-generated data and tested on real prospective data showed performance comparable to a model trained solely on real data.
- These results indicate that synthetic data can maintain the predictive power of AI models while enhancing privacy.
Conclusions:
- Synthetic radiomics data, particularly generated by BGMMOCE, offers a viable solution for privacy preservation in AI-driven healthcare applications.
- This approach successfully balances data fidelity and privacy, enabling robust AI models for prostate cancer aggressiveness classification.
- The findings support the use of synthetic data to mitigate privacy risks associated with radiomics in clinical AI.
