Parsing disease heterogeneity in structural and functional MRI-derived measures using normative modeling and...

Area of Science:

• Neuroimaging
• Artificial Intelligence in Medicine
• Computational Neuroscience

Background:

• Neurological disorders exhibit significant heterogeneity, complicating diagnosis and treatment.
• Existing normative modeling techniques often require large, disease-specific datasets, limiting their applicability.
• Individual-level deviations in brain structure and function are key indicators of neurological conditions.

Purpose of the Study:

• To develop and validate a Generative Adversarial Network (GAN)-based normative modeling technique for capturing individual brain measure deviations.
• To address the challenge of data scarcity for rare or heterogeneous neurological disorders.
• To demonstrate the method's utility in identifying neuroanatomical and functional abnormalities in specific diseases.

Main Methods:

• Utilized a GAN framework for self-supervised training on pseudo-synthetically simulated patient data.
• Applied the normative modeling technique to structural Magnetic Resonance Imaging (MRI) and resting-state functional MRI (fMRI) data.
• Evaluated the model's ability to detect disease-related effects without large, annotated datasets.

Main Results:

• The GAN-based model successfully identified individual-level deviations in brain measures.
• Neuroanatomical deviations were detected in structural MRI data for Alzheimer's disease (AD) and Traumatic Brain Injury (TBI).
• Functional connectivity abnormalities were identified in resting-state fMRI data for AD and TBI.

Conclusions:

• The proposed GAN-based normative modeling is a versatile and effective tool for detecting disease-related brain abnormalities.
• This approach facilitates the study of individual deviations in neurological disorders, even with limited disease-specific data.
• The method shows promise for personalized diagnosis and advancing research in brain disorders.