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

Generalisation capabilities of machine-learning algorithms for the detection of the subthalamic nucleus in micro-electrode recordings

PURPOSE

Micro-electrode recordings (MERs) are a key intra-operative modality used during deep brain stimulation (DBS) electrode implantation, which allow for a trained neurophysiologist to infer the anatomy in which the electrode is placed. As DBS targets are small, such inference is necessary to confirm that the electrode is correctly positioned. Recently, machine learning techniques have been used to augment the neurophysiologist's capability. The goal of this paper is to investigate the generalisability of these methods with respect to different clinical centres and training paradigms.

METHODS

Five deep learning algorithms for binary classification of MER signals have been implemented. Three databases from two different clinical centres have also been collected with differing size, acquisition hardware, and annotation protocol. Each algorithm has initially been trained on the largest database, then either directly tested or fine-tuned on the smaller databases in order to estimate their generalisability. As a reference, they have also been trained from scratch on the smaller databases as well in order to estimate the effect of the differing database sizes and annotation systems.

RESULTS

Each network shows significantly reduced performance (on the order of a 6.5% to 16.0% reduction in balanced accuracy) when applied out-of-distribution. This reduction can be ameliorated through fine-tuning the network on the new database through transfer learning. Although, even for these small databases, it appears that retraining from scratch may still offer equivalent performance as fine-tuning with transfer learning. However, this is at the expense of significantly longer training times.

CONCLUSION

Generalisability is an important criterion for the success of machine learning algorithms in clinic. We have demonstrated that a variety of recent machine learning algorithms for MER classification are negatively affected by domain shift, but that this can be quickly ameliorated through simple transfer learning procedures that can be readily performed for new centres.

A Systematic Review of Neurophysiology-Based Localization Techniques Used in Deep Brain Stimulation Surgery of the Subthalamic Nucleus

OBJECTIVE

This systematic review is conducted to identify, compare, and analyze neurophysiological feature selection, extraction, and classification to provide a comprehensive reference on neurophysiology-based subthalamic nucleus (STN) localization.

MATERIALS AND METHODS

The review was carried out using the methods and guidelines of the Kitchenham systematic review and provides an in-depth analysis on methods proposed on STN localization discussed in the literature between 2000 and 2021. Three research questions were formulated, and 115 publications were identified to answer the questions.

RESULTS

The three research questions formulated are answered using the literature found on the respective topics. This review discussed the technologies used in past research, and the performance of the state-of-the-art techniques is also reviewed.

CONCLUSION

This systematic review provides a comprehensive reference on neurophysiology-based STN localization by reviewing the research questions other new researchers may also have.

Localization of deep brain stimulation trajectories via automatic mapping of microelectrode recordings to MRI

Objective. Suboptimal electrode placement during subthalamic nucleus deep brain stimulation (STN DBS) surgery may arise from several sources, including frame-based targeting errors and intraoperative brain shift. We present a computer algorithm that can accurately localize intraoperative microelectrode recording (MER) tracks on preoperative magnetic resonance imaging (MRI) in real-time, thereby predicting deviation between the surgical plan and the MER trajectories.Approach. Random forest (RF) modeling was used to derive a statistical relationship between electrophysiological features on intraoperative MER and voxel intensity on preoperative T2-weighted MR imaging. This model was integrated into a larger algorithm that can automatically localize intraoperative MER recording tracks on preoperative MRI in real-time. To verify accuracy, targeting error of both the planned intraoperative trajectory ('planned') and the algorithm-derived trajectory ('calculated') was estimated by measuring deviation from the final DBS lead location on postoperative high-resolution computed tomography ('actual').Main results. MR imaging and MERs were obtained from 24 STN DBS implant trajectories. The cross-validated RF model could accurately distinguish between gray and white matter regions along MER trajectories (AUC 0.84). When applying this model within the localization algorithm, thecalculatedMER trajectory estimate was found to be significantly closer to theactualDBS lead when compared to theplannedtrajectory recorded during surgery (1.04 mm vs 1.52 mm deviation,p< 0.002), with improvement shown in 19/24 cases (79%). When applying the algorithm to simulated DBS trajectory plans with randomized targeting error, up to 4 mm of error could be resolved to <2 mm on average (p< 0.0001).Significance. This work presents an automated system for intraoperative localization of electrodes during STN DBS surgery. This neuroengineering solution may enhance the accuracy of electrode position estimation, particularly in cases where high-resolution intraoperative imaging is not available.

Deep brain stimulation for Parkinson’s Disease: A Review and Future Outlook

Parkinson's Disease (PD) is a neurodegenerative disorder that manifests as an impairment of motor and non-motor abilities due to a loss of dopamine input to deep brain structures. While there is presently no cure for PD, a variety of pharmacological and surgical therapeutic interventions have been developed to manage PD symptoms. This review explores the past, present and future outlooks of PD treatment, with particular attention paid to deep brain stimulation (DBS), the surgical procedure to deliver DBS, and its limitations. Finally, our group's efforts with respect to brain mapping for DBS targeting will be discussed.

Machine learning in deep brain stimulation: A systematic review

Deep Brain Stimulation (DBS) is an increasingly common therapy for a large range of neurological disorders, such as abnormal movement disorders. The effectiveness of DBS in terms of controlling patient symptomatology has made this procedure increasingly used over the past few decades. Concurrently, the popularity of Machine Learning (ML), a subfield of artificial intelligence, has skyrocketed and its influence has more recently extended to medical domains such as neurosurgery. Despite its growing research interest, there has yet to be a literature review specifically on the use of ML in DBS. We have followed a fully systematic methodology to obtain a corpus of 73 papers. In each paper, we identified the clinical application, the type/amount of data used, the method employed, and the validation strategy, further decomposed into 12 different sub-categories. The papers overall illustrated some existing trends in how ML is used in the context of DBS, including the breath of the problem domain and evolving techniques, as well as common frameworks and limitations. This systematic review analyzes at a broad level how ML have been recently used to address clinical problems on DBS, giving insight into how these new computational methods are helping to push the state-of-the-art of functional neurosurgery. DBS clinical workflow is complex, involves many specialists, and raises several clinical issues which have partly been addressed with artificial intelligence. However, several areas remain and those that have been recently addressed with ML are by no means considered "solved" by the community nor are they closed to new and evolving methods.

Artificial Intelligence Technologies in Neurosurgery: a Systematic Literature Review Using Topic Modeling. Part II: Research Objectives and Perspectives

The current increase in the number of publications on the use of artificial intelligence (AI) technologies in neurosurgery indicates a new trend in clinical neuroscience.

The aim of the study was to conduct a systematic literature review to highlight the main directions and trends in the use of AI in neurosurgery.

Automatic identification of sweet spots from MERs for electrodes implantation in STN-DBS

PURPOSE

Microelectrode recordings (MERs) are a significant clinical indicator for sweet spots identification of implanted electrodes during deep brain stimulation of the subthalamic nucleus (STN) surgery. As 1D MERs signals have the unboundedness, large-range, large-amount and time-dependent characteristics, the purpose of this study is to propose an automatic and precise identification method of sweet spots from MERs, reducing the time-consuming and labor-intensive human annotations.

METHODS

We propose an automatic identification method of sweet spots from MERs for electrodes implantation in STN-DBS. To better imitate the surgeons' observation and obtain more intuitive contextual information, we first employ the 2D Gramian angular summation field (GASF) images generated from MERs data to perform the sweet spots determination for electrodes implantation. Then, we introduce the convolutional block attention module into convolutional neural network (CNN) to identify the 2D GASF images of sweet spots for electrodes implantation.

RESULTS

Experimental results illustrate that the identification result of our method is consistent with the result of doctor's decision, while our method can achieve the accuracy and precision of 96.72% and 98.97%, respectively, which outperforms state-of-the-art for intraoperative sweet spots determination.

CONCLUSIONS

The proposed method is the first time to automatically and accurately identify sweet spots from MERs for electrodes implantation by the combination an advanced time series-to-image encoding way with CBAM-enhanced networks model. Our method can assist neurosurgeons in automatically detecting the most likely locations of sweet spots for electrodes implantation, which can provide an important indicator for target selection while it reduces the localization error of the target during STN-DBS surgery.

Deep convolutional neural network for the automated detection of Subthalamic nucleus using MER signals

BACKGROUND

Deep brain stimulation (DBS) surgery has been extensively conducted for treating advanced Parkinson's disease (PD) patient's symptoms. DBS hinges on the localization of the subthalamic nucleus (STN) in which a permanent electrode should be accurately placed to produce electrical current. Microelectrode recording (MER) signals are routinely recorded in the procedure of DBS surgery to validate the planned trajectories. However, manual MER signals interpretation with the goal of detecting STN borders requires expertise and prone to inter-observer variability. Therefore, a computerized aided system would be beneficial to automatic detection of the dorsal and ventral borders of the STN in MER.

NEW METHOD

In this study, a new deep learning model based on convolutional neural system for automatic delineation of the neurophysiological borders of the STN along the electrode trajectory was developed.

COMPARISON WITH EXISTING METHODS

The proposed model does not involve any conventional standardization, feature extraction or selection steps.

RESULTS

Promising results of 98.67% accuracy, 99.03% sensitivity, 98.11% specificity, 98.79% precision and 98.91% F1-score for subject based testing were achieved using the proposed convolutional neural network (CNN) model.

CONCLUSIONS

This is the first study on the analysis of MER signals to detect STN using deep CNN. Traditional machine learning (ML) algorithms are often cumbersome and suffer from subjective evaluation. Though, the developed 10-layered CNN model has the capability of extracting substantial features at the convolution stage. Hence, the proposed model has the potential to deliver high performance on STN region detection which shows perspective in aiding the neurosurgeon intraoperatively.

An intra-operative feature-based classification of microelectrode recordings to support the subthalamic nucleus functional identification during deep brain stimulation surgery

OBJECTIVE

The Subthalamic Nucleus (STN) is the most selected target for the placement of the Deep Brain Stimulation (DBS) electrode to treat Parkinson's disease. Its identification is a delicate and challenging task which is based on the interpretation of the STN functional activity acquired through microelectrode recordings (MER). Aim of this work is to explore the potentiality of a set of twenty-five features to build a classification model for the discrimination of MER signals belonging to the STN.

APPROACH

We explored the use of different sets of spike-dependent and spike-independent features in combination with an Ensemble Trees classification (ET) algorithm on a dataset composed of thirteen patients receiving bilateral DBS. We compared results from six subsets of features and two dataset conditions (with and without standardization) using performance metrics on a leave-one-patient-out validation schema.

MAIN RESULTS

We obtained statistically better results (i.e., higher accuracy p-value = 0.003) on the raw dataset than on the standardized one, where the selection of seven features using a minimum redundancy maximum relevance (MRMR) algorithm provided a mean accuracy of 94.1%, comparable with the use of the full set of features. In the same conditions, the spike-dependent features provided the lowest accuracy (86.8%), while a power density-based index was shown to be a good indicator of STN activity (92.3%).

SIGNIFICANCE

Results suggest that a small and simple set of features can be used for an efficient classification of microelectrode recordings to implement an intraoperative support for clinical decision during deep brain stimulation surgery.

SepaConvNet for Localizing the Subthalamic Nucleus Using One Second Micro-electrode Recordings

Micro-electrode recording (MER) is a powerful way of localizing target structures during neurosurgical procedures such as the implantation of deep brain stimulation electrodes, which is a common treatment for Parkinson's disease and other neurological disorders. While Micro-electrode Recording (MER) provides adjunctive information to guidance assisted by pre-operative imaging, it is not unanimously used in the operating room. The lack of standard use of MER may be in part due to its long duration, which can lead to complications during the operation, or due to high degree of expertise required for their interpretation. Over the past decade, various approaches addressing automating MER analysis for target localization have been proposed, which have mainly focused on feature engineering. While the accuracies obtained are acceptable in certain configurations, one issue with handcrafted MER features is that they do not necessarily capture more subtle differences in MER that could be detected auditorily by an expert neurophysiologist. In this paper, we propose and validate a deep learning-based pipeline for subthalamic nucleus (STN) localization with micro-electrode recordings motivated by the human auditory system. Our proposed Convolutional Neural Network (CNN), referred as SepaConvNet, shows improved accuracy over two comparative networks for locating the STN from one second MER samples.

Decoding kinetic features of hand motor preparation from single-trial EEG using convolutional neural networks

Building accurate movement decoding models from brain signals is crucial for many biomedical applications. Predicting specific movement features, such as speed and force, before movement execution may provide additional useful information at the expense of increasing the complexity of the decoding problem. Recent attempts to predict movement speed and force from the electroencephalogram (EEG) achieved classification accuracies at or slightly above chance levels, highlighting the need for more accurate prediction strategies. Thus, the aims of this study were to accurately predict hand movement speed and force from single-trial EEG signals and to decode neurophysiological information of motor preparation from the prediction strategies. To these ends, a decoding model based on convolutional neural networks (ConvNets) was implemented and compared against other state-of-the-art prediction strategies, such as support vector machines and decision trees. ConvNets outperformed the other prediction strategies, achieving an overall accuracy of 84% in the classification of two different levels of speed and force (four-class classification) from pre-movement single-trial EEG (100 ms and up to 1,600 ms prior to movement execution). Furthermore, an analysis of the ConvNet architectures suggests that the network performs a complex spatiotemporal integration of EEG data to optimize classification accuracy. These results show that movement speed and force can be accurately predicted from single-trial EEG, and that the prediction strategies may provide useful neurophysiological information about motor preparation.

Intraoperative Localization of STN During DBS Surgery Using a Data-Driven Model

A new approach is presented for localizing the Subthalamic Nucleus (STN) during Deep Brain Stimulation (DBS) surgery based on microelectrode recordings (MERs). DBS is an accepted treatment for individuals living with Parkinson’s Disease (PD). This surgery involves implantation of a permanent electrode inside the STN to deliver electrical current. Since the STN is a very small region inside the brain, accurate placement of an electrode is a challenging task for the surgical team. Prior to placement of the permanent electrode, microelectrode recordings of brain activity are used intraoperatively to localize the STN. The placement of the electrode and the success of the therapy depend on this location. In this paper, an objective approach is implemented to help the surgical team in localizing the STN. This is achieved by processing the MER signals and extracting features during the surgery to be used in a Machine Learning (ML) algorithm for defining the neurophysiological borders of the STN. For this purpose, a new classification approach is proposed with the goal of detecting both the dorsal and the ventral borders of the STN during the surgical procedure. Results collected from 100 PD patients in this study, show that by calculating and extracting wavelet transformation features from MER signals and using a data-driven computational deep neural network model, it is possible to detect the borders of the STN with an accuracy of 92%. The proposed method can be implemented in real-time during the surgery to model the neurophysiological nonlinearity along the path of the electrode trajectory during insertion.

Automatic feature group combination selection method based on GA for the functional regions clustering in DBS

BACKGROUND AND OBJECTIVE

The functional regions clustering through microelectrode recording (MER) is a critical step in deep brain stimulation (DBS) surgery. The localization of the optimal target highly relies on the neurosurgeon's empirical assessment of the neurophysiological signal. This work presents an unsupervised clustering algorithm to get the optimal cluster result of the functional regions along the electrode trajectory.

METHODS

The dataset consists of the MERs obtained from the routine bilateral DBS for PD patients. Several features have been extracted from MER and divided into groups based on the type of neurophysiological signal. We selected single feature groups rather than all features as the input samples of each division of the divisive hierarchical clustering (DHC) algorithm. And the optimal cluster result has been achieved through a feature group combination selection (FGS) method based on genetic algorithm (GA). To measure the performance of this method, we compared the accuracy and validation indexes of three methods, including DHC only, DHC with GA-based FGS and DHC with GA-based feature selection (FS) in other studies, on the universal and DBS datasets.

RESULTS

Results show that the DHC with GA-based FGS achieved the optimal cluster result compared with other methods. The three borders of the STN can be identified from the cluster result. The dorsoventral sizes of the STN and dorsal STN are 4.45 mm and 2.02 mm. In addition, the features extracted from the multiunit activity, background unit activity and local field potential are found to be the most representative feature groups to identify the dorsal, d-v and ventral borders of the STN, respectively.

CONCLUSIONS

Our clustering algorithm showed a reliable performance of the automatic identification of functional regions in DBS. The findings can provide valuable assistance for both neurosurgeons and stereotactic surgical robots in DBS surgery.

Optimization of the KNN Supervised Classification Algorithm as a Support Tool for the Implantation of Deep Brain Stimulators in Patients with Parkinson's Disease

Deep Brain Stimulation (DBS) of the Subthalamic Nuclei (STN) is the most used surgical treatment to improve motor skills in patients with Parkinson’s Disease (PD) who do not adequately respond to pharmacological treatment, or have related side effects. During surgery for the implantation of a DBS system, signals are obtained through microelectrodes recordings (MER) at different depths of the brain. These signals are analyzed by neurophysiologists to detect the entry and exit of the STN region, as well as the optimal depth for electrode implantation. In the present work, a classification model is developed and supervised by the K-nearest neighbour algorithm (KNN), which is automatically trained from the 18 temporal features of MER registers of 14 patients with PD in order to provide a clinical support tool during DBS surgery. We investigate the effect of different standardizations of the generated database, the optimal definition of KNN configuration parameters, and the selection of features that maximize KNN performance. The results indicated that KNN trained with data that was standardized per cerebral hemisphere and per patient presented the best performance, achieving an accuracy of 94.35% (p < 0.001). By using feature selection algorithms, it was possible to achieve 93.5% in accuracy in selecting a subset of six features, improving computation time while processing in real time.

Realtime phase-amplitude coupling analysis of micro electrode recorded brain signals

Objective

To demonstrate a method to calculate phase amplitude coupling (PAC) quickly and robustly for realtime applications.

Methods

We designed and implemented a multirate PAC algorithm with efficient filter bank processing and efficient computation of PAC for many frequency-pair combinations. We tested the developed algorithm for computing PAC on simulated data and on intraoperative neural recording data obtained during deep brain stimulation (DBS) electrode implantation surgery.

Results

A combination of parallelized frequency-domain filtering and modulation index for PAC estimation provided robust results that could be calculated in real time on modest computing hardware.

Conclusion

The standard methods for calculating PAC can be optimized for quick and robust performance.

Significance

These results demonstrated that PAC can be extracted in real time and is suitable for neurofeedback applications.

Classification of pallidal oscillations with increasing parkinsonian severity

The firing patterns of neurons in the basal ganglia are known to become more oscillatory and synchronized from healthy to parkinsonian conditions. Similar changes have been observed with local field potentials (LFPs). In this study, we used an unbiased machine learning approach to investigate the utility of pallidal LFPs for discriminating the stages of a progressive parkinsonian model. A feature selection algorithm was used to identify subsets of LFP features that provided the most discriminatory information for severity of parkinsonian motor signs. Prediction errors <20% were achievable using 28 of the possible 206 features tested. For all subjects, a spectral feature within the beta band was chosen through the feature selection algorithm, but a combination of features, including alpha-band power and phase-amplitude coupling, was necessary to achieve minimal prediction errors. There was large variability between the discriminatory features for individual subjects, and testing of classifiers between subjects yielded prediction errors >50%. These results suggest that pallidal oscillations can be predictive biomarkers of parkinsonian severity, but the features are more complex than spectral power in individual frequency bands, such as the beta band. Additionally, the best feature set was subject specific, which highlights the pathophysiological heterogeneity of parkinsonism and the importance of subject specificity when designing closed-loop system controllers dependent on such features.