Statistical Validation of Synthetic Data for Lung Cancer Patients Generated by Using Generative Adversarial Networks

Abstract

The development of healthcare patient digital twins in combination with machine learning technologies helps doctors in therapeutic prescription and in minimally invasive intervention procedures. The confidentiality of medical records or limited data availability in many health domains are drawbacks that can be overcome with the generation of synthetic data conformed to real data. The use of generative adversarial networks (GAN) for the generation of synthetic data of lung cancer patients has been previously introduced as a tool to solve this problem in the form of anonymized synthetic patients. However, generated synthetic data are mainly validated from the machine learning domain (loss functions) or expert domain (oncologists). In this paper, we propose statistical decision making as a validation tool: Is the model good enough to be used? Does the model pass rigorous hypothesis testing criteria? We show for the case at hand how loss functions and hypothesis validation are not always well aligned.