Supplementary Table S4: Machine learning assisting neurosurgical care of Parkinson’s Disease patients

Authors / Year / Title / Journal / Stage within neurosurgical care / Algorithm / Input features / Application / Results/Conclusion
Volume Estimation
Chaturvedi, A., Lujan, J. L. and McIntyre, C. C. / 2013 / Artificial neural network based characterization of the volume of tissue activated during deep brain stimulation / J Neural Eng / Intraoperative Assistance / ANN / MERs / Estimate volume of activated tissue during DBS / The ANN-based predictor provides more accurate descriptions of the spatial spread of activation compared to activation function-based approaches for monopolar stimulation. In addition, the ANN was able to accurately estimate the volume of tissue activated in response to multi-contact electrode configurations.
Identify eloquent area
Nowinski, W. L., Belov, D. and Benabid, A. L. / 2003 / An algorithm for rapid calculation of a probabilistic functional atlas of subcortical structures from electrophysiological data collected during functional neurosurgery procedures / Neuroimage / Intraoperative Assistance / GMM / MERs / Identify eloquent area / It contains the most popular stereotactic targets, the subthalamic nucleus, globus pallidus internus, and ventral intermedius nucleus. The key application of the algorithm is targeting in stereotactic and functional neurosurgery, and it also can be employed in human and animal brain research.
Intra-operative localization
Shamir, R., Duchin, Y., Kim, J. J. K., Marmor, O., Bergman, H., Vitek, J. L., Sapiro, G., Bick, A. S., Eliyahu, R., Eitan, R., Israel, Z. and Harel, N. / 2016 / MER validation of a new targeting approach for STN-DBS surgery based on machine-learning and 7T-MRI database (10661) / Neuromodulation / Intraoperative Assistance / ANN / MERs / Localize subthalamic nucleus during DBS / 100% (20/20) of the electrodes intersect the predicted subthalamic nucleus as confirmed with the MER. 89% (71/80) of the contacts were classified the same by the MER and our prediction. The observed accuracies per individual contact are: 70% (14/20), 95% (19/20), 100% (20/20), and 90% (18/20) for contacts 3, 2, 1, and 0, respectively ('0' is the most ventral contact).
Telkes, I., Jimenez-Shahed, J., Viswanathan, A., Abosch, A. and Ince, N. F. / 2016 / Prediction of STN-DBS electrode implantation track in Parkinson's disease by using local field potentials / Frontiers in Neuroscience / Intraoperative Assistance / QDA / MERs / Localize subthalamic nucleus during DBS / By fusing the information from these low and high frequency bands, the dorsal border of subthalamic nucleus was localized with a root mean square error of 1.22 mm. The prediction accuracy for the optimal track was 80%. Individual beta band (11-32 Hz) and the range of high frequency oscillations (200-450 Hz) provided prediction accuracies of 72 and 68% respectively. The best prediction result obtained with monopolar local field potentials data was 68%.
Teplitzky, B. A., Zitella, L. M., Xiao, Y. and Johnson, M. D. / 2016 / Model-Based Comparison of Deep Brain Stimulation Array Functionality with Varying Number of Radial Electrodes and Machine Learning Feature Sets / Front Comput Neurosci / Intraoperative Assistance / Multiple (ANN, BL, RF, KNN, SVM) / MERs / Localize subthalamic nucleus during DBS / The results of this study indicate that, for a clinical-scale DBS lead, more than four radial electrodes minimally improved in the ability to steer, shift, and sculpt axonal activation around a DBS lead and a simple feature set consisting of the RoA center of mass and orientation enabled robust machine learning classification.
Rajpurohit, V., Danish, S. F., Hargreaves, E. L. and Wong, S. / 2015 / Optimizing computational feature sets for subthalamic nucleus localization in DBS surgery with feature selection / Clin Neurophysiol / Intraoperative Assistance / Multiple (BL, GMM, KNN, LR, SVM) / MERs / Localize subthalamic nucleus during DBS / Feature selection, patient-specific normalization, and both produced relative error reductions of 4.95%, 31.36%, and 38.92%, respectively. Three of four feature-selected logistic refression classifiers performed better than 99% of classifiers with all possible feature combinations. Optimal feature combinations were not predictable from individual feature performance.
Szlufik, S., Ciecierski, K., Rola, R., Mandat, T., Nauman, P., Ras, Z., Przybyszewski, A., Friedman, A. and Koziorowski, D. / 2015 / DBS decision support system based on analysis of microelectrode recorded signals as an useful tool in detection of the most beneficial electrode localization during DBS implantation / Brain Stimulation / Intraoperative Assistance / ANN / MERs / Localize subthalamic nucleus during DBS / On the basis of analysis of the power of electrodes used in patients during microrecording from subthalamic nucleus, the sensitivity of DBS-DSS was estimated on 84,7% (left hemisphere mean value - 88,1%, right hemisphere - 80% ) and the specificity was calculated to be at the level of 95,9% (left hemisphere mean value - 95,6%, right hemisphere - 96,1% ). Even higher sensitivity and specificity values were obtained in the cross validation of the classifier
Teplitzky, B. A., Xiao, J., Zitella, L. and Johnson, M. D. / 2014 / Design considerations and guided programming of deep brain stimulation arrays using computational modeling / Neuromodulation / Intraoperative Assistance / SVM / MERs / Localize subthalamic nucleus during DBS / Classification crossvalidation resulted in a 2%false-negative rate and 0% false-positive rate. The two-part algorithm converged in 100% of test-cases and required less than 30 seconds of computation time using desktop machine.
Taghva, A. / 2011 / Hidden semi-Markov models in the computerized decoding of microelectrode recording data for deep brain stimulator placement / World Neurosurgery / Intraoperative Assistance / BL / MERs / Localize subthalamic nucleus during DBS / Accuracy of these competing models was similar for correctly identifying brain nuclei; however, the algorithm showed superior specificity in detecting microelectrode passes traversing the subthalamic nucleus.
Taghva, A. / 2010 / An automated navigation system for deep brain stimulator placement using hidden Markov models / Neurosurgery / Intraoperative Assistance / BL / MERs / Localize subthalamic nucleus during DBS / The accuracy of identifying the correct brain location was 98.5%. Sensitivity of detecting passes intersecting the subthalamic nucleus was 100%, and specificity was 84.9%. Anatomical location of the MER passes was calculated with a mean error of 0.06 mm (95% confidence interval, -0.54 to 0.42 mm) in the medial-lateral axis.
Wong, S., Baltuch, G. H., Jaggi, J. L. and Danish, S. F. / 2009 / Functional localization and visualization of the subthalamic nucleus from microelectrode recordings acquired during DBS surgery with unsupervised machine learning / J Neural Eng / Intraoperative Assistance / FCM / MERs / Localize subthalamic nucleus during DBS / Using these maps, a non-neurophysiologist is able to achieve sensitivities of 90% and 95% for subthalamic nucleus entry and exit, respectively, to within 0.5 mm accuracy of the current gold standard.
Zaidel, A., Spivak, A., Shpigelman, L., Bergman, H. and Israel, Z. / 2009 / Delimiting subterritories of the human subthalamic nucleus by means of microelectrode recordings and a Hidden Markov Model / Mov Disord / Intraoperative Assistance / BL / MERs / Localize subthalamic nucleus during DBS / The algorithm identified subthalamic nucleus-entry, the ventral boundary of the dorsolateral oscillatory region, and subthalamic nucleus-exit with an error of -0.09 +/- 0.35, -0.27 +/- 0.58, and -0.20 +/- 0.33 mm, respectively (mean +/- standard deviation), and with detection reliability (error < 1 mm) of 95, 86, and 91%, respectively.
Moran, A., Bar-Gad, I., Bergman, H. and Israel, Z. / 2006 / Real-time refinement of subthalamic nucleus targeting using Bayesian decision-making on the root mean square measure / Mov Disord / Intraoperative Assistance / BL / MERs / Localize subthalamic nucleus during DBS / We tested our predictions on each trajectory using a bootstrapping technique, with the rest of the trajectories serving as a training set and found the error in predicting the subthalamic nucleus entry to be (mean +/- SD) 0.18 +/- 0.84, and 0.50 +/- 0.59 mm for subthalamic nucleus exit point, which yields a 0.30 +/- 0.28 mm deviation from the expert's target center.
Predict symptom improvement
Kostoglou, K., Michmizos, K. P., Stathis, P., Sakas, D., Nikita, K. S. and Mitsis, G. D. / 2016 / Classification and Prediction of Clinical Improvement in Deep Brain Stimulation from Intraoperative Microelectrode Recordings / IEEE Trans Biomed Eng / Neurosurgical outcome prediction / RF / MRI / Predict symptom improvement after surgery / We modified the employed RFs to account for unbalanced data sets and multiple observations per patient, and showed, for the first time, that only 5 neurophysiologically interpretable MER signal features are sufficient for predicting UPDRS improvement.
Shamir, R. R., Dolber, T., Noecker, A. M., Walter, B. L. and McIntyre, C. C. / 2015 / Machine Learning Approach to Optimizing Combined Stimulation and Medication Therapies for Parkinson's Disease / Brain Stimul / Neurosurgical outcome prediction / Multiple (BL, RF, SVM) / Clinical / Predict symptom improvement after surgery / Using those results, the combined machine learning algorithms were able to accurately predict 86% (12/14) of the motor improvement scores at one year after surgery.
Segmentation
Fu, Y., Gao, W., Zhu, M., Chen, X., Lin, Z. and Wang, S. / 2009 / Computer-assisted automatic localization of the human pedunculopontine nucleus in T1-weighted MR images: a preliminary study / Int J Med Robot / Presurgical Planning / ANN / MRI / Segmentation of critical/target brain structure / Experiments were performed with both T1-weighted and proton density MR images acquired from 12 people. Preliminary results show that the proposed method can locate the prepontine nucleus with an error of 1.83 +/- 0.42 mm for its rostral pole and 1.57 +/- 0.34 mm for its caudal pole.

Abbreviations: ANN: Artificial Neural Networks; BCI: Brain; Computer Interface; BL: Bayesion Learning; CT: Computer Tomography; DBS: Deep Brain Stimulation; DT: Decision Tree; EEG: Electro-EncephaloGraphy; FCM: Fuzzy C-Means; FIS: Fuzzy Inference System; GA: Genetic Algorithm; GB: Gradient Boosting; GMM: Gaussian Mixture Models; ICP: Intracranial Pressue; iEEG: Intracranial Electro-EncephaloGraphy; KNN: K-Nearest Neighbors; LDA: Linear Discriminant Analysis; LR: Logistic Regression; MERs: Micro-Electrode Recordings; MRI: Magnetic Resonance Imaging; NLP: Natural language processing; OLS: Ordinary Least Squares; PCA: Principal Component Analysis; PET: Positrion Emission Tomography; QDA:Quadratic Discriminant Analysis