[Machine learning for resting state fMRI-based preoperative mapping: comparison with task-based fMRI and direct...

Area of Science:

• Neurosurgery
• Neuroimaging
• Machine Learning

Background:

• Brain gliomas pose challenges for preserving critical motor and speech functions during surgery.
• Accurate preoperative mapping of eloquent brain areas is crucial for minimizing neurological deficits.

Purpose of the Study:

• To develop and validate a machine learning system for predicting individual brain activations in motor and speech areas.
• To compare the efficacy of resting-state fMRI (rs-fMRI) and task-based fMRI (tb-fMRI) for preoperative mapping in glioma patients.

Main Methods:

• A machine learning system was pre-trained on 200 healthy subjects' data from The Human Connectome Project.
• The system was further trained and validated using preoperative MRI (T1, tb-fMRI, rs-fMRI) data from 33 glioma patients.
• Direct cortical stimulation (DCS) was used intraoperatively to verify functional area predictions.

Main Results:

• rs-fMRI predicted more positive direct cortical stimulation responses (132) compared to tb-fMRI (112).
• tb-fMRI showed a higher proportion of negative stimulation sites within activated areas (69 vs. 44).

Conclusions:

• The developed machine learning method using rs-fMRI demonstrated superior sensitivity in identifying speech zones (0.72 vs. 0.66) and motor areas (0.79 vs. 0.62) compared to tb-fMRI.
• rs-fMRI-based prediction, verified by DCS, offers a more sensitive approach for preoperative functional mapping in glioma surgery.