Related Experiment Video
Updated: Jul 7, 2025
14:27
Identification of Disease-related Spatial Covariance Patterns using Neuroimaging Data
Published on: June 26, 2013
15.7K
Predicting disease-related MRI patterns of multiple sclerosis through GAN-based image editing
Daniel Güllmar1, Wei-Chan Hsu1, Jürgen R Reichenbach1
1Medical Physics Group, Institute for Diagnostic and Interventional Radiology, University Hospital Jena, Jena 07743, Germany; Michael Stifel Center for Data-Driven and Simulation Science, Jena 07743, Germany.
Zeitschrift Fur Medizinische Physik
|December 24, 2023
Summary
Deep learning with StyleGAN can simulate multiple sclerosis (MS) progression using MRI scans. This approach reveals patterns of brain atrophy and lesion development, aiding in understanding MS disease markers.
Area of Science:
- Neuroimaging
- Artificial Intelligence in Medicine
- Deep Learning
Background:
- Multiple sclerosis (MS) is a complex neurodegenerative disorder affecting the central nervous system.
- Magnetic resonance imaging (MRI) is crucial for diagnosing and monitoring MS.
- Current methods for predicting MS progression have limitations.
Purpose of the Study:
- To apply a deep learning model (StyleGAN) for simulating MS progression in MRI.
- To explore image markers associated with MS using generative adversarial networks.
- To investigate the relationship between brain atrophy and MS progression.
Main Methods:
- Unsupervised training of StyleGAN using T1-weighted GRE MRI and ADC maps from MS patients and controls.
- Latent space manipulation of MR images to simulate disease progression.
- Analysis of simulated progression by comparing intensity profiles and calculating brain parenchymal fraction (BPF).
Main Results:
- StyleGAN successfully simulated MS progression through latent space manipulation.
- Simulated progression showed brain volume loss in T1-weighted and ADC maps.
- Increasing lesion extent was observed in simulated ADC maps.
Conclusions:
- The StyleGAN model shows potential for studying MS image markers.
- Latent space manipulation can mimic MS disease progression and atrophy.
- This deep learning approach offers new insights into MS pathophysiology.
