Optimizing computational feature sets for subthalamic nucleus localization in DBS surgery with feature selection.

Area of Science:

• Neurosurgery
• Computational Neuroscience
• Biomedical Engineering

Background:

• Microelectrode recording (MER) is crucial for identifying the subthalamic nucleus (STN) during deep brain stimulation (DBS) surgery.
• Automated STN detection relies on extracting quantitative features from MER data for classifier training.

Purpose of the Study:

• To evaluate the efficacy of feature selection in identifying optimal feature combinations for automated STN localization.
• To compare the performance of feature-selected classifiers against those using all possible feature combinations.

Main Methods:

• Extracted 13 features from 65 MERs for logistic regression (LR) classification.
• Employed hold-one-patient-out cross-validation to assess classification error.
• Compared patient-specific versus independent normalization strategies.

Main Results:

• Feature selection combined with patient-specific normalization yielded the greatest relative error reduction (38.92%).
• Feature selection alone reduced classification error by 4.95%, while patient-specific normalization reduced it by 31.36%.
• Four feature-selected LR classifiers outperformed 99% of classifiers using all features.

Conclusions:

• Feature selection effectively reduces classification error in automated STN localization from MER data.
• Patient-specific normalization further enhances accuracy, indicating its necessity for clinically reliable MER interpretation.
• These advancements contribute to the automated functional localization of the STN in DBS surgery.