Multi-patient learning increases accuracy for Subthalamic Nucleus identification in deep brain stimulation.

Area of Science:

• Neurosurgery
• Machine Learning
• Computational Neuroscience

Background:

• Accurate localization of basal ganglia is critical for brain surgeries like deep brain stimulation (DBS) for Parkinson's disease.
• Current automated systems often use patient-independent classifiers trained on diverse patient data.
• Improving the precision of targeting specific brain regions remains an ongoing challenge.

Purpose of the Study:

• To investigate the efficacy of a multi-task learning framework for enhancing the accuracy of Subthalamic Nucleus targeting.
• To demonstrate the benefits of sharing information across different patient datasets in surgical localization tasks.
• To compare the performance of the proposed multi-task approach against traditional patient-independent methods.

Main Methods:

• Implementation of a multi-task learning framework to simultaneously train on related tasks.
• Utilizing shared information from multiple patient datasets to improve classifier performance.
• Evaluation of the system's accuracy on two distinct real-world datasets.

Main Results:

• The multi-task learning framework significantly increased the accuracy of Subthalamic Nucleus targeting compared to patient-independent methods.
• Performance gains were validated across two independent real-world datasets.
• Demonstrated the effectiveness of leveraging cross-patient data through multi-task learning.

Conclusions:

• Multi-task learning offers a superior approach for patient-independent localization in neurosurgical applications.
• Sharing information across patients enhances the precision of targeting critical brain structures like the Subthalamic Nucleus.
• This framework holds promise for improving the outcomes of deep brain stimulation surgery.