Generative artificial intelligence model for simulating structural brain changes in schizophrenia.

Area of Science:

• Neuroimaging
• Artificial Intelligence
• Psychiatric Disorders

Background:

• Generative AI offers new methods for creating realistic medical images, enhancing patient privacy.
• AI-driven image generation can augment limited neuroimaging datasets for training discriminative models.
• The application of generative AI for simulating complex psychiatric disorders is largely unexplored.

Purpose of the Study:

• To develop a novel generative AI model for transforming healthy MRI images into those resembling schizophrenia (SZ).
• To explore the application of this model in simulating psychiatric conditions and disease progression.

Main Methods:

• Utilized anonymized MRI datasets from the Center for Biomedical Research Excellence and Autism Brain Imaging Data Exchange.
• Developed a cycle generative adversarial network (cGAN) model to transform healthy subject (HS) MRI images into SZ-like images.
• Evaluated transformation efficacy using voxel-based morphometry and age prediction accuracy; assessed simulation of comorbidities and disease progression.

Main Results:

• The AI model successfully transformed HS images into SZ-like images, reflecting known brain volume changes.
• Simulations highlighted structural differences in comorbidities like autism spectrum disorder (ASD).
• The model demonstrated realistic simulation of disease progression while preserving individual characteristics.

Conclusions:

• The generative AI model effectively captures subtle brain changes associated with schizophrenia.
• This tool provides novel visualization of disease-related brain alterations.
• The model has potential applications in simulating disease mechanisms and refining therapeutic strategies.