Feature selection for KNN classifier to improve accurate detection of subthalamic nucleus during deep brain stimulation surgery in Parkinson’s patients

Abstract

The tremor and dystonia associated with Parkinson’s disease can be treated with deep brain stimulation (DBS) implanted into the subthalamic nucleus (STN). The accurate STN detection is a complex neurosurgeon task during a DBS surgery since a proper fixing of stimulating electrodes will impact on the patient’s future life. The brain electrical signals obtained with Micro Electrodes Register (MER) are acquired at different depths of the brain during DBS surgery to detect STN. In our previous work, we found good accuracy performance to improve the localization of STN using K-Nearest Neighbours (KNN) supervised learning algorithm. However, for real-time classification, it is essential to reduce the feature dimension without loss of classification accuracy so as to reduce computation time. In this study, we compared the KNN classification trained with 16 features with other KNN with a reduced number of features resulting from applying four feature selection techniques. We obtained similar classification results compared to the classifier using all features, with 6 feature resulting from Branch & Bound (KNN+B) algorithm, giving the best performancewith 86.13% accuracy and 92.72% AUC. From the study, we infer that the KNN+B classifier has the potential to be used in detecting the STN in real time