Aditi Sathe

Student(s): Aditi Sathe

Project: Graph Neural Networks in Medical Imagining | View Poster (PDF)

Major(s): Bioengineering (Data Science minor)

Advisor(s): Akum Onwunta

Abstract

Graph Neural Networks (GNNs) have emerged as a powerful tool in machine learning research, proving particularly effective for analyzing structured data such as graphs. This capability has shown promising potential in the application of analyzing neural images. Neural imaging datasets, inherently structured as intricate networks where nodes represent different regions of the brain and edges signify neural connections, can be profoundly analyzed using GNNs to dynamically understand brain functionality and structure. The primary advantage of using GNNs in neural imaging lies in their ability to directly process the complex, graph-like data architecture. This capability enables the preservation of the topological structure of neural networks, which is crucial for understanding inter-regional brain activity and connectivity. By employing techniques like message passing, GNNs facilitate the aggregation of information across connected nodes (brain regions), allowing for the analysis of localized patterns within broader network interactions. Specifically, GNNs have been particularly useful in studies analyzing fMRI (functional MRI) data, capturing both the localized activation of specific brain regions and intricate interactions between different areas over time. Additionally, GNNs assist in exploring changes in connectivity patterns over time, providing insights into how brain network dynamics correlate with behavioral outputs or disease states. The versatility of GNNs extends beyond functional analysis; they are also adept at integrating multimodal data sources, such as combining fMRI with structural MRI (sMRI) or EEG data. This multimodal approach enhances the accuracy and robustness of brain network analyses, leading to better predictive models for neurological diseases.