Automated eloquent cortex localization in brain tumor patients using multi-task graph neural networks.

Area of Science:

• Neuroscience
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate localization of the eloquent cortex is vital for effective presurgical planning.
• Task-based functional MRI (fMRI) is often challenging for surgical patients.
• Resting-state fMRI (rs-fMRI) offers a noninvasive alternative, but automated localization remains difficult.

Purpose of the Study:

• To develop a novel deep learning architecture for simultaneous identification of language and primary motor cortex from rs-fMRI connectivity.
• To address the challenge of automated eloquent cortex localization using noninvasive neuroimaging.

Main Methods:

• A novel deep learning architecture combining convolutional neural networks and multi-task learning was developed.
• The model identifies language and primary motor cortex from rs-fMRI connectivity data.
• Validation was performed on datasets from the Human Connectome Project and a brain tumor cohort.

Main Results:

• The proposed deep learning model achieved significantly better performance compared to traditional machine learning and fully-connected deep learning baselines.
• The graph convolution architecture with multi-task learning demonstrated superior results in localizing eloquent cortical areas.
• The method showed generalizability and robustness across different datasets.

Conclusions:

• The developed deep learning approach effectively localizes the eloquent cortex using rs-fMRI.
• This noninvasive method shows significant promise for improving presurgical planning.
• The findings highlight the potential of rs-fMRI combined with advanced AI for clinical neurosurgery.