Data-Driven Prediction of the Therapeutic Window during Subthalamic Deep Brain Stimulation Surgery

Abstract

Background: Moving from awake surgery under local anesthesia to asleep surgery under general anesthesia will require to precisely predict the outcome of deep brain stimulation. Objective: To propose a data-driven prediction of both the therapeutic effect and side effects of the surgery. Methods: The retrospective intraoperative data from 30 patients operated on in the subthalamic nucleus were used to train an artificial neural network to predict the deep brain stimulation outcome. A leave-one-out validation was undertaken to give a predictive performance that would reflect the performance of the predictive model in clinical practice. Three-dimensional coordinates and the amount of current of the electrodes were used to train the model. Results: 130 electrode positions were reviewed. The areas under the curve were 0.902 and 0.89 for therapeutic and side effects, respectively. The mean sensitivity and specificity were 93.07% (SD 0.95) and 69.24% (SD 5.27) for the therapeutic effect, 73.47% (SD 10.55) and 91.82% (SD 0.12) for the side effect. Conclusion: Data-driven prediction could be an additional modality to predict deep brain stimulation outcome. Further validation is needed to precisely use this method for performing surgery under general anesthesia.

Survival prediction of glioblastoma patients—are we there yet? A systematic review of prognostic modeling for glioblastoma and its clinical potential

“…The increasing use of machine learning for the analysis of unstructured, high-dimensional data parallels the current trends in predictive modeling in medicine [44,45]. Neurosurgical examples include machine learning algorithms for glioblastoma, deep brain stimulation, traumatic brain injury, stroke, and spine surgery [38,[44][45][46][47][48][49][50][51][52][53]. Deep learning algorithms are also increasingly being used to further improve the WHO 2016 classification of high-grade gliomas via histological and biomolecular variables for more concise diagnosis and classification of gliomas [54][55][56].…”

Machine learning in deep brain stimulation: A systematic review

“…Baumgarten et al [25,29,36] and Bermudez et al [58] proposed clinical efficacy probability maps to visualize the expected clinical effects of stimulation of several locations around the structure of interest. Singer et al [57] went even further in this idea by directly predicting the optimal electrode location without the use of intermediate representations such as anatomical segmentations or clinical-effect probability maps.…”

“…Therefore, CNNs learn an optimal dimensionality reduction strategy in a supervised manner. Among the other occurrences are the three papers from Baumgarten et al [25,29,36] and the paper from Habets et al [55], because the number of inputs is low (respectively four stimulation parameters and 15 clinical scores and demographics).…”

“…However, there were some papers that simply did not separate training and testing patients, could lead to biased measurements of their performance. The results reported by these contributions are interesting as preliminary work, but cannot safely be considered as representative of a real, prospective usage.Others measured their performance by employing a patient-wise validation method[11,12,21,22,26,27,32,29,33,36,47,48,51,54,55,57,58,61,62,63,64,65,66,69,70,72,73,74,75,76,77,78,79,80,81,82]. While it does not ensure the complete absence of data leakage (that could occur, for example, by mixing the validation and testing sets, or by selecting or normalize features with the whole database), these results can be considered as more reliable, and more representative of prospective-usage performance.…”