A Novel Metric for Detecting Memorization in Generative Models for Brain MRI Synthesis
Antonio Scardace*,
Lemuel Puglisi*,
Francesco Guarnera,
Sebastiano Battiato, and
Daniele Ravì
* Joint first authors
This repository contains the official code for our paper on DeepSSIM, a novel automated metric for detecting training data memorization in generative models applied to medical imaging. While generative models show great promise in generating synthetic medical data, they are vulnerable to memorizing sensitive information. DeepSSIM addresses this risk by detecting and quantifying memorized content.
We evaluate DeepSSIM in a case study involving synthetic Brain MRI data generated by a Latent Diffusion Model (LDM) trained under memorization-prone conditions, using 2,195 2D MRI scans from two publicly available datasets - IXI and CoRR. Compared to state-of-the-art memorization metrics, DeepSSIM achieves superior performance, improving F1 score by an average of +52.03% over the best existing method.
Instructions will be uploaded upon acceptance.
Instructions will be uploaded upon acceptance.
Instructions will be uploaded upon acceptance.
Instructions will be uploaded upon acceptance.
Please, reference this publication if you find this code useful:
[wip]
