Machine Learning Approach to Optimizing Combined Stimulation and Medication Therapies for Parkinson's Disease

An Artificial Neural Network Predicts Gender Differences of Motor and Non-Motor Symptoms of Patients with Advanced Parkinson's Disease under Levodopa-Carbidopa Intestinal Gel

Background and Objectives: Currently, no tool exists to predict clinical outcomes in patients with advanced Parkinson's disease (PD) under levodopa-carbidopa intestinal gel (LCIG) treatment. The aim of this study was to develop a novel deep neural network model to predict the clinical outcomes of patients with advanced PD after two years of LCIG therapy. Materials and Methods: This was a longitudinal, 24-month observational study of 59 patients with advanced PD in a multicenter registry under LCIG treatment from September 2019 to September 2021, including 43 movement disorder centers. The data set includes 649 measurements of patients, which make an irregular time series, and they are turned into regular time series during the preprocessing phase. Motor status was assessed with the Unified Parkinson's Disease Rating Scale (UPDRS) Parts III (off) and IV. The NMS was assessed by the NMS Questionnaire (NMSQ) and the Geriatric Depression Scale (GDS), the quality of life by PDQ-39, and severity by Hoehn and Yahr (HY). Multivariate linear regression, ARIMA, SARIMA, and Long Short-Term Memory-Recurrent NeuralNetwork (LSTM-RNN) models were used. Results: LCIG significantly improved dyskinesia duration and quality of life, with men experiencing a 19% and women a 10% greater improvement, respectively. Multivariate linear regression models showed that UPDRS-III decreased by 1.5 and 4.39 units per one-unit increase in the PDQ-39 and UPDRS-IV indexes, respectively. Although the ARIMA-(2,0,2) model is the best one with AIC criterion 101.8 and validation criteria MAE = 0.25, RMSE = 0.59, and RS = 0.49, it failed to predict PD patients' features over a long period of time. Among all the time series models, the LSTM-RNN model predicts these clinical characteristics with the highest accuracy (MAE = 0.057, RMSE = 0.079, RS = 0.0053, mean square error = 0.0069). Conclusions: The LSTM-RNN model predicts, with the highest accuracy, gender-dependent clinical outcomes in patients with advanced PD after two years of LCIG therapy.

Machine learning for adaptive deep brain stimulation in Parkinson's disease: closing the loop

Parkinson's disease (PD) is the second most common neurodegenerative disease bearing a severe social and economic impact. So far, there is no known disease modifying therapy and the current available treatments are symptom oriented. Deep Brain Stimulation (DBS) is established as an effective treatment for PD, however current systems lag behind today's technological potential. Adaptive DBS, where stimulation parameters depend on the patient's physiological state, emerges as an important step towards "smart" DBS, a strategy that enables adaptive stimulation and personalized therapy. This new strategy is facilitated by currently available neurotechnologies allowing the simultaneous monitoring of multiple signals, providing relevant physiological information. Advanced computational models and analytical methods are an important tool to explore the richness of the available data and identify signal properties to close the loop in DBS. To tackle this challenge, machine learning (ML) methods applied to DBS have gained popularity due to their ability to make good predictions in the presence of multiple variables and subtle patterns. ML based approaches are being explored at different fronts such as the identification of electrophysiological biomarkers and the development of personalized control systems, leading to effective symptom relief. In this review, we explore how ML can help overcome the challenges in the development of closed-loop DBS, particularly its role in the search for effective electrophysiology biomarkers. Promising results demonstrate ML potential for supporting a new generation of adaptive DBS, with better management of stimulation delivery, resulting in more efficient and patient-tailored treatments.

Perspective of artificial intelligence in healthcare data management: A journey towards precision medicine

Mounting evidence has highlighted the implementation of big data handling and management in the healthcare industry to improve the clinical services. Various private and public companies have generated, stored, and analyzed different types of big healthcare data, such as omics data, clinical data, electronic health records, personal health records, and sensing data with the aim to move in the direction of precision medicine. Additionally, with the advancement in technologies, researchers are curious to extract the potential involvement of artificial intelligence and machine learning on big healthcare data to enhance the quality of patient's lives. However, seeking solutions from big healthcare data requires proper management, storage, and analysis, which imposes hinderances associated with big data handling. Herein, we briefly discuss the implication of big data handling and the role of artificial intelligence in precision medicine. Further, we also highlighted the potential of artificial intelligence in integrating and analyzing the big data that offer personalized treatment. In addition, we briefly discuss the applications of artificial intelligence in personalized treatment, especially in neurological diseases. Lastly, we discuss the challenges and limitations imposed by artificial intelligence in big data management and analysis to hinder precision medicine.

Early and High-Accuracy Diagnosis of Parkinson's Disease: Outcomes of a New Model

Parkinson's disease (PD) is one of the significant common neurological disorders of the current age that causes uncontrollable movements like shaking, stiffness, and difficulty. The early clinical diagnosis of this disease is essential for preventing the progression of PD. Hence, an innovative method is proposed here based on combining the crow search algorithm and decision tree (CSADT) for the early PD diagnosis. This approach is used on four crucial Parkinson's datasets, including meander, spiral, voice, and speech-Sakar. Using the presented method, PD is effectively diagnosed by evaluating each dataset's critical features and extracting the primary practical outcomes. The used algorithm was compared with other machine learning algorithms of k-nearest neighbor (KNN), support vector machine (SVM), naive Baye (NB), multilayer perceptron (MLP), decision tree (DT), random tree, logistic regression, support vector machine of radial base functions (SVM of RBFs), and combined classifier in terms of accuracy, recall, and combination measure F1. The analytical results emphasize the used algorithm's superiority over the other selected ones. The proposed model yields nearly 100% accuracy through various trials on the datasets. Notably, a high detection speed achieved the lowest detection time of 2.6 seconds. The main novelty of this paper is attributed to the accuracy of the presented PD diagnosis method, which is much higher than its counterparts.

Explainable machine learning models based on multimodal time-series data for the early detection of Parkinson's disease

BACKGROUND AND OBJECTIVES

Parkinson's Disease (PD) is a devastating chronic neurological condition. Machine learning (ML) techniques have been used in the early prediction of PD progression. Fusion of heterogeneous data modalities proved its capability to improve the performance of ML models. Time series data fusion supports the tracking of the disease over time. In addition, the trustworthiness of the resulting models is improved by adding model explainability features. The literature on PD has not sufficiently explored these three points.

METHODS

In this work, we proposed an ML pipeline for predicting the progression of PD that is both accurate and explainable. We explore the fusion of different combinations of five time series modalities from the Parkinson's Progression Markers Initiative (PPMI) real-world dataset, including patient characteristics, biosamples, medication history, motor, and non-motor function data. Each patient has six visits. The problem has been formulated in two ways: ❶ a three-class based progression prediction with 953 patients in each time series modality, and ❷ a four-class based progression prediction with 1,060 patients in each time series modality. The statistical features of these six visits were calculated from each modality and diverse feature selection methods were applied to select the most informative feature sets. The extracted features were used to train a set of well-known ML models including Support vector machines (SVM), random forests (RF), extra tree classifier (ETC), light gradient boosting machines (LGBM), and stochastic gradient descent (SGD). We examined a number of data-balancing strategies in the pipeline with different combinations of modalities. ML models have been optimized using the Bayesian optimizer. A comprehensive evaluation of various ML methods has been conducted, and the best models have been extended to provide different explainability features.

RESULTS

We compare the performance of ML models before and after optimization and using and without using feature selection. In the three-class experiment and with various modality fusions, the LGBM model produced the most accurate results with a 10-fold cross-validation (10-CV) accuracy of 90.73% using non-motor function modality. RF produced the best results in the four-class experiment with various modality fusions with a 10-CV accuracy of 94.57% using non-motor modality. With the fused dataset of non-motor and motor function modalities, the LGBM model outperformed the other ML models in both the 3-class and 4-class experiments (i.e., 10-CV accuracy of 94.89% and 93.73%, respectively). Using the Shapely Additive Explanations (SHAP) framework, we employed global and instance-based explanations to explain the behavior of each ML classifier. Moreover, we extended the explainability by implementing the LIME and SHAPASH local explainers. The consistency of these explainers has been explored. The resultant classifiers were accurate, explainable, and thus medically more relevant and applicable.

CONCLUSIONS

The select modalities and feature sets were confirmed by the literature and medical experts. The various explainers suggest that the bradykinesia (NP3BRADY) feature was the most dominant and consistent. By providing thorough insights into the influence of multiple modalities on the disease risk, the suggested approach is expected to help improve the clinical knowledge of PD progression processes.

A Combined Method for Diabetes Mellitus Diagnosis Using Deep Learning, Singular Value Decomposition, and Self-Organizing Map Approaches

Diabetes in humans is a rapidly expanding chronic disease and a major crisis in modern societies. The classification of diabetics is a challenging and important procedure that allows the interpretation of diabetic data and diagnosis. Missing values in datasets can impact the prediction accuracy of the methods for the diagnosis. Due to this, a variety of machine learning techniques has been studied in the past. This research has developed a new method using machine learning techniques for diabetes risk prediction. The method was developed through the use of clustering and prediction learning techniques. The method uses Singular Value Decomposition for missing value predictions, a Self-Organizing Map for clustering the data, STEPDISC for feature selection, and an ensemble of Deep Belief Network classifiers for diabetes mellitus prediction. The performance of the proposed method is compared with the previous prediction methods developed by machine learning techniques. The results reveal that the deployed method can accurately predict diabetes mellitus for a set of real-world datasets.

A Practical Statin Recommendation System Based on Real-World Data to Improve LDL-C Management in Secondary Prevention

ABSTRACT

Statins are considered the cornerstone of secondary prevention in patients with atherosclerotic cardiovascular disease (ASCVD). However, many patients fail to achieve the guide-recommended goal of low-density lipoprotein cholesterol (LDL-C) after statin monotherapy, leading to a high residual risk of cardiovascular events. Owing to individual differences in statin therapy, it is possible first to consider changing the type of statin before adding nonstatin medications in certain patients to improve LDL-C management. We developed and evaluated a statin recommendation system using real-world data. Ensemble learning was performed to develop the recommendation system that integrated the output results of support vector machines (SVM) and the similarity of patients. Model performance was assessed to investigate whether treatment according to the recommended model would increase the proportion of patients with the primary end point. Finally, a total of 3510 patients were enrolled in the development and validation of the recommender system. Of them, 1240 patients received atorvastatin (35.3%), 1714 patients received rosuvastatin (48.8%), and 556 patients received pitavastatin (15.8%). The statin recommendation system could significantly improve LDL-C target rate achievement in the recommended treatment group compared with the nonrecommended treatment group in the validation set (50.8% vs. 31.5%, P < 0.001). This study demonstrated that the statin recommendation system could significantly improve the achievement of LDL-C goals in ASCVD patients, providing a new approach to improve LDL-C management.

Discovering Themes in Deep Brain Stimulation Research Using Explainable Artificial Intelligence

Deep brain stimulation is a treatment that controls symptoms by changing brain activity. The complexity of how to best treat brain dysfunction with deep brain stimulation has spawned research into artificial intelligence approaches. Machine learning is a subset of artificial intelligence that uses computers to learn patterns in data and has many healthcare applications, such as an aid in diagnosis, personalized medicine, and clinical decision support. Yet, how machine learning models make decisions is often opaque. The spirit of explainable artificial intelligence is to use machine learning models that produce interpretable solutions. Here, we use topic modeling to synthesize recent literature on explainable artificial intelligence approaches to extracting domain knowledge from machine learning models relevant to deep brain stimulation. The results show that patient classification (i.e., diagnostic models, precision medicine) is the most common problem in deep brain stimulation studies that employ explainable artificial intelligence. Other topics concern attempts to optimize stimulation strategies and the importance of explainable methods. Overall, this review supports the potential for artificial intelligence to revolutionize deep brain stimulation by personalizing stimulation protocols and adapting stimulation in real time.

Determination of the motor status of patients with advanced Parkinson’s disease under levodopa–carbidopa intestinal gel using a machine learning model

BACKGROUND

Despite the careful selection of candidate patients, the levodopa-carbidopa intestinal gel (LCIG) treatment of advanced Parkinson's disease (PD) remains challenging due to a complex interplay between motor and non-motor symptoms. We developed a random forest (RF) model to determine the postoperative motor outcome of patients with advanced PD at 2 years under the LCIG therapy by using motor and non-motor data from a Greek multicenter, observational registry (ForHealth S.A.).

METHODS

This was a prospective 24-month, observational study of 59 patients with advanced PD under LCIG treatment from September 2019 to September 2021. Motor status was assessed with the Unified Parkinson's Disease Rating Scale (UPDRS) parts III and IV. Non-motor symptoms (NMS) were assessed by the Non-Motor Symptoms Questionnaire (NMSQ) and the Geriatric Depression Scale (GDS).

RESULTS

We demonstrated that the proper combination of motor and non-motor measures significantly determines the motor outcome (UPDRS-III year 2: 23.57 ± 14.22 p < 0.001), reducing the RMSE (root-mean-square-error) from 3.487279 to 3.066292, suggesting that the optimized model performed well. Based on the "IncNodePurity," the major determinant factors of UPDRS-III (year 2) were, in descending order: UPDRS-III (year 0), disease duration, NMSQ (year 2), age, NMSQ (year 0), time off (hours) (year 2), time dyskinesia (year 0), quality of life (year 2) after the LCIG implementation.

CONCLUSIONS

The novelty of this model is the possibility to determine the motor outcome after two years of LCIG. This model could be also useful for not specialized Parkinson's neurologists, to improve patient counseling, expectation management, and patient satisfaction with LCIG therapy.

Deep brain-machine interfaces: sensing and modulating the human deep brain

Different from conventional brain-machine interfaces that focus more on decoding the cerebral cortex, deep brain-machine interfaces enable interactions between external machines and deep brain structures. They sense and modulate deep brain neural activities, aiming at function restoration, device control and therapeutic improvements. In this article, we provide an overview of multiple deep brain recording and stimulation techniques that can serve as deep brain-machine interfaces. We highlight two widely used interface technologies, namely deep brain stimulation and stereotactic electroencephalography, for technical trends, clinical applications and brain connectivity research. We discuss the potential to develop closed-loop deep brain-machine interfaces and achieve more effective and applicable systems for the treatment of neurological and psychiatric disorders.

Subthalamic nucleus deep brain stimulation programming settings do not correlate with Parkinson's disease severity

BACKGROUND

Deep brain stimulation (DBS) is a well-established treatment for Parkinson's disease (PD). While the success of DBS is dependent on careful patient selection and accurate lead placement, programming parameters play a pivotal role in tailoring therapy on the individual level. Various algorithms have been developed to streamline the initial programming process, but the relationship between pre-operative patient characteristics and post-operative device settings is unclear. In this study, we investigated how PD severity correlates with DBS settings.

METHODS

We conducted a retrospective review of PD patients who underwent DBS of the subthalamic nucleus at one US tertiary care center between 2014 and 2018. Pre-operative patient characteristics and post-operative programming data at various intervals were collected. Disease severity was measured using the Unified Parkinson's Disease Rating Scale score (UPDRS) as well as levodopa equivalent dose (LED). Correlation analyses were conducted looking for associations between pre-operative disease severity and post-operative programming parameters.

RESULTS

Fifty-six patients were analyzed. There was no correlation between disease severity and any of the corresponding programming parameters. Pre-operative UPDRS scores on medication were similar to post-operative scores with DBS. Settings of amplitude, frequency, and pulse width increased significantly from 1 to 6 months post-operatively. Stimulation volume, inferred by the distance between contacts used, also increased significantly over time.

CONCLUSIONS

Interestingly, we found that patients with more advanced disease responded to electrical stimulation similarly to patients with less advanced disease. These data provide foundational knowledge of DBS programming parameters used in a single cohort of PD patients over time.

Artificial intelligence applications and robotic systems in Parkinson's disease (Review)

Parkinson's disease (PD) is the second most frequent neurodegenerative disorder following Alzheimer's disease. Advanced stages of PD, 4 or 5 of the Hoehn and Yahr Scale, are characterized by severe motor complications, limited mobility without assistance, risk of falling, and non-motor complications. The aim of this review was to provide a practical overview on specific artificial intelligence (AI) systems for the management of advanced stages of PD, as well as relevant technological limitations. The authors conducted a systematic search on PubMed and EMBASE with predefined keywords searching for studies published until December 2020. Full articles that satisfied the inclusion criteria were included in the systematic review. To minimize results bias, the reference list was manually searched for pertinent articles to identify any additional relevant missed publications. Exclusion criteria included the following: Other stages of PD than 4 and 5 of the Hoehn and Yahr Scale, case reports, reviews, practice guidelines, commentaries, opinions, letters, editorials, short surveys, articles in press, conference abstracts, conference papers, and abstracts published without a full article. The search identified 21 studies analyzing AI-based applications and robotic systems used for the management of advanced stages of PD, out of which 6 articles analyzed AI-based applications for autonomous management of pharmacologic therapy, 5 articles analyzed home-based telemedicine systems and 10 articles analysed robot-assisted gait training systems. The authors identified significant evidence demonstrating that current AI-based technologies are feasible for automatic management of patients with advanced stages of PD. Improving the quality of care and reducing the cost for patients and healthcare systems are the most important advantages.

Personalizing decision-making for persons with Parkinson's disease: where do we stand and what to improve?

Background

The large variety in symptoms and treatment effects across different persons with Parkinson’s disease (PD) warrants a personalized approach, ensuring that the best decision is made for each individual. We aimed to further clarify this process of personalized decision-making, from the perspective of medical professionals.

Methods

We audio-taped 52 consultations with PD patients and their neurologist or PD nurse-specialist, in 6 outpatient clinics. We focused coding of the transcripts on which decisions were made and on if and how decisions were personalized. We subsequently interviewed professionals to elaborate on how and why decisions were personalized, and which decisions would benefit most from a more personalized approach.

Results

Most decisions were related to medication, referral or lifestyle. Professionals balanced clinical factors, including individual (disease-) characteristics, and non-clinical factors, including patients’ preference, for each type of decision. These factors were often not explicitly discussed with the patient. Professionals experienced difficulties in personalizing decisions, mostly because evidence on the impact of characteristics of an individual patient on the outcome of the decision is unavailable. Categories of decisions for which professionals emphasized the importance of a more personalized perspective include choices not only for medication and advanced treatments, but also for referrals, lifestyle and diagnosis.

Conclusions

Clinical decision-making is a complex process, influenced by many different factors that differ for each decision and for each individual. In daily practice, it proves difficult to tailor decisions to individual (disease-) characteristics, probably because sufficient evidence on the impact of these individual characteristics on outcomes is lacking.

Predict initial subthalamic nucleus stimulation outcome in Parkinson's disease with brain morphology

AIM

Subthalamic nucleus deep brain stimulation (STN-DBS) has been reported to be effective in treating motor symptoms in Parkinson's disease (PD), which may be attributed to changes in the brain network. However, the association between brain morphology and initial STN-DBS efficacy, as well as the performance of prediction using neuroimaging, has not been well illustrated. Therefore, we aim to investigate these issues.

METHODS

In the present study, 94 PD patients underwent bilateral STN-DBS, and the initial stimulation efficacy was evaluated. Brain morphology was examined by magnetic resonance imaging (MRI). The volume of tissue activated in the motor STN was measured with MRI and computed tomography. The prediction of stimulation efficacy was achieved with a support vector machine, using brain morphology and other features, after feature selection and hyperparameter optimization.

RESULTS

A higher stimulation efficacy was correlated with a thicker right precentral cortex. No association with subcortical gray or white matter volumes was observed. These morphological features could estimate the individual stimulation response with an r value of 0.5678, an R² of 0.3224, and an average error of 11.4%. The permutation test suggested these predictions were not based on chance.

CONCLUSION

Our results indicate that changes in morphology are associated with the initial stimulation motor response and could be used to predict individual initial stimulation-related motor responses.

Foundations of Bayesian Learning in Clinical Neuroscience

There is an increasing interest in using prediction models to forecast clinical outcomes within the fields of neurosurgery and clinical neuroscience. The present chapter outlines the foundations of Bayesian learning and introduces Bayes theorem and its use in machine learning methodology. The use of Bayesian networks to structure and define associations between outcome predictors and final outcomes is highlighted and Naïve Bayes classifiers are outlined for use in predicting neurosurgical outcomes, where the understanding of underlying causes is less important. The present work aims to orient researchers in Bayesian machine learning methods and when and how to use them. When used correctly, these tools have the potential to improve the understanding of factors influencing neurosurgical outcomes, aid in structuring the relationships between them, and provide reliable machine learning classification models for predicting neurosurgical outcomes.

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.

Quadruple Decision Making for Parkinson's Disease Patients: Combining Expert Opinion, Patient Preferences, Scientific Evidence, and Big Data Approaches to Reach Precision Medicine

Clinical decision making for Parkinson’s disease patients is supported by a combination of three distinct information resources: best available scientific evidence, professional expertise, and the personal needs and preferences of patients. All three sources have clear value but also share several important limitations, mainly regarding subjectivity, generalizability and variability. For example, current scientific evidence, especially from controlled clinical trials, is often based on selected study populations, making it difficult to translate the outcome to the care for individual patients in everyday clinical practice. Big data, including data from real-life unselected Parkinson populations, can help to bridge this information gap. Fine-grained patient profiles created from big data have the potential to aid in identifying therapeutic approaches that will be most effective given each patient’s individual characteristics, which is particularly important for a disorder characterized by such tremendous interindividual variability as Parkinson’s disease. In this viewpoint, we argue that big data approaches should be acknowledged and harnessed, not to replace existing information resources, but rather as a fourth and complimentary source of information in clinical decision making, helping to represent the full complexity of individual patients. We introduce the ‘quadruple decision making’ model and illustrate its mode of action by showing how this can be used to pursue precision medicine for persons living with Parkinson’s disease.

Probabilistic Mapping of Deep Brain Stimulation: Insights from 15 Years of Therapy

Deep brain stimulation (DBS) depends on precise delivery of electrical current to target tissues. However, the specific brain structures responsible for best outcome are still debated. We applied probabilistic stimulation mapping to a retrospective, multidisorder DBS dataset assembled over 15 years at our institution (n_total = 482 patients; n_{Parkinson disease} = 303; n_dystonia = 64; n_tremor = 39; n_{treatment-resistant depression/anorexia nervosa} = 76) to identify the neuroanatomical substrates of optimal clinical response. Using high-resolution structural magnetic resonance imaging and activation volume modeling, probabilistic stimulation maps (PSMs) that delineated areas of above-mean and below-mean response for each patient cohort were generated and defined in terms of their relationships with surrounding anatomical structures. Our results show that overlap between PSMs and individual patients' activation volumes can serve as a guide to predict clinical outcomes, but that this is not the sole determinant of response. In the future, individualized models that incorporate advancements in mapping techniques with patient-specific clinical variables will likely contribute to the optimization of DBS target selection and improved outcomes for patients. ANN NEUROL 2021;89:426-443.

Machine Learning's Application in Deep Brain Stimulation for Parkinson's Disease: A Review

Deep brain stimulation (DBS) is a surgical treatment for advanced Parkinson's disease (PD) that has undergone technological evolution that parallels an expansion in clinical phenotyping, neurophysiology, and neuroimaging of the disease state. Machine learning (ML) has been successfully used in a wide range of healthcare problems, including DBS. As computational power increases and more data become available, the application of ML in DBS is expected to grow. We review the literature of ML in DBS and discuss future opportunities for such applications. Specifically, we perform a comprehensive review of the literature from PubMed, the Institute for Scientific Information's Web of Science, Cochrane Database of Systematic Reviews, and Institute of Electrical and Electronics Engineers' (IEEE) Xplore Digital Library for ML applications in DBS. These studies are broadly placed in the following categories: (1) DBS candidate selection; (2) programming optimization; (3) surgical targeting; and (4) insights into DBS mechanisms. For each category, we provide and contextualize the current body of research and discuss potential future directions for the application of ML in DBS.

Artificial Intelligence in Pharmacoepidemiology: A Systematic Review. Part 1-Overview of Knowledge Discovery Techniques in Artificial Intelligence

Aim

To perform a systematic review on the application of artificial intelligence (AI) based knowledge discovery techniques in pharmacoepidemiology.

Study Eligibility Criteria

Clinical trials, meta-analyses, narrative/systematic review, and observational studies using (or mentioning articles using) artificial intelligence techniques were eligible. Articles without a full text available in the English language were excluded.

Data Sources

Articles recorded from 1950/01/01 to 2019/05/06 in Ovid MEDLINE were screened.

Participants

Studies including humans (real or simulated) exposed to a drug.

Results

In total, 72 original articles and 5 reviews were identified via Ovid MEDLINE. Twenty different knowledge discovery methods were identified, mainly from the area of machine learning (66/72; 91.7%). Classification/regression (44/72; 61.1%), classification/regression + model optimization (13/72; 18.0%), and classification/regression + features selection (12/72; 16.7%) were the three most frequent tasks in reviewed literature that machine learning methods has been applied to solve. The top three used techniques were artificial neural networks, random forest, and support vector machines models.

Conclusions

The use of knowledge discovery techniques of artificial intelligence techniques has increased exponentially over the years covering numerous sub-topics of pharmacoepidemiology.

Systematic Review Registration

Systematic review registration number in PROSPERO: CRD42019136552.

Homecare Robotic Systems for Healthcare 4.0: Visions and Enabling Technologies

Powered by the technologies that have originated from manufacturing, the fourth revolution of healthcare technologies is happening (Healthcare 4.0). As an example of such revolution, new generation homecare robotic systems (HRS) based on the cyber-physical systems (CPS) with higher speed and more intelligent execution are emerging. In this article, the new visions and features of the CPS-based HRS are proposed. The latest progress in related enabling technologies is reviewed, including artificial intelligence, sensing fundamentals, materials and machines, cloud computing and communication, as well as motion capture and mapping. Finally, the future perspectives of the CPS-based HRS and the technical challenges faced in each technical area are discussed.

Modulation of Nigrofugal and Pallidofugal Pathways in Deep Brain Stimulation for Parkinson Disease

BACKGROUND

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is a well-established surgical therapy for patients with Parkinson disease (PD).

OBJECTIVE

To define the role of adjacent white matter stimulation in the effectiveness of STN-DBS.

METHODS

We retrospectively evaluated 43 patients with PD who received bilateral STN-DBS. The volumes of activated tissue were analyzed to obtain significant stimulation clusters predictive of 4 clinical outcomes: improvements in bradykinesia, rigidity, tremor, and reduction of dopaminergic medication. Tractography of the nigrofugal and pallidofugal pathways was performed. The significant clusters were used to calculate the involvement of the nigrofugal and pallidofugal pathways and the STN.

RESULTS

The clusters predictive of rigidity and tremor improvement were dorsal to the STN with most of the clusters outside of the STN. These clusters preferentially involved the pallidofugal pathways. The cluster predictive of bradykinesia improvement was located in the central part of the STN with an extension outside of the STN. The cluster predictive of dopaminergic medication reduction was located ventrolateral and caudal to the STN. These clusters preferentially involved the nigrofugal pathways.

CONCLUSION

Improvements in rigidity and tremor mainly involved the pallidofugal pathways dorsal to the STN. Improvement in bradykinesia mainly involved the central part of the STN and the nigrofugal pathways ventrolateral to the STN. Maximal reduction in dopaminergic medication following STN-DBS was associated with an exclusive involvement of the nigrofugal pathways.

Brain morphological changes in hypokinetic dysarthria of Parkinson's disease and use of machine learning to predict severity

Background

Up to 90% of patients with Parkinson's disease (PD) eventually develop the speech and voice disorder referred to as hypokinetic dysarthria (HD). However, the brain morphological changes associated with HD have not been investigated. Moreover, no reliable model for predicting the severity of HD based on neuroimaging has yet been developed.

Methods

A total of 134 PD patients were included in this study and divided into a training set and a test set. All participants underwent a structural magnetic resonance imaging (MRI) scan and neuropsychological evaluation. Individual cortical thickness, subcortical structure, and white matter volume were extracted, and their association with HD severity was analyzed. After feature selection, a machine‐learning model was established using a support vector machine in the training set. The severity of HD was then predicted in the test set.

Results

Atrophy of the right precentral cortex and the right fusiform gyrus was significantly associated with HD. No association was found between HD and volume of white matter or subcortical structures. Favorable and optimal performance of machine learning on HD severity prediction was achieved using feature selection, giving a correlation coefficient (r) of .7516 and a coefficient of determination (R²) of .5649 (P < .001).

Conclusion

The brain morphological changes were associated with HD. Excellent prediction of the severity of HD was achieved using machine learning based on neuroimaging.

Microelectrode Recordings Validate the Clinical Visualization of Subthalamic-Nucleus Based on 7T Magnetic Resonance Imaging and Machine Learning for Deep Brain Stimulation Surgery

BACKGROUND

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is a proven and effective therapy for the management of the motor symptoms of Parkinson's disease (PD). While accurate positioning of the stimulating electrode is critical for success of this therapy, precise identification of the STN based on imaging can be challenging. We developed a method to accurately visualize the STN on a standard clinical magnetic resonance imaging (MRI). The method incorporates a database of 7-Tesla (T) MRIs of PD patients together with machine-learning methods (hereafter 7 T-ML).

OBJECTIVE

To validate the clinical application accuracy of the 7 T-ML method by comparing it with identification of the STN based on intraoperative microelectrode recordings.

METHODS

Sixteen PD patients who underwent microelectrode-recordings guided STN DBS were included in this study (30 implanted leads and electrode trajectories). The length of the STN along the electrode trajectory and the position of its contacts to dorsal, inside, or ventral to the STN were compared using microelectrode-recordings and the 7 T-ML method computed based on the patient's clinical 3T MRI.

RESULTS

All 30 electrode trajectories that intersected the STN based on microelectrode-recordings, also intersected it when visualized with the 7 T-ML method. STN trajectory average length was 6.2 ± 0.7 mm based on microelectrode recordings and 5.8 ± 0.9 mm for the 7 T-ML method. We observed a 93% agreement regarding contact location between the microelectrode-recordings and the 7 T-ML method.

CONCLUSION

The 7 T-ML method is highly consistent with microelectrode-recordings data. This method provides a reliable and accurate patient-specific prediction for targeting the STN.

How Big Data and High-performance Computing Drive Brain Science

Brain science accelerates the study of intelligence and behavior, contributes fundamental insights into human cognition, and offers prospective treatments for brain disease. Faced with the challenges posed by imaging technologies and deep learning computational models, big data and high-performance computing (HPC) play essential roles in studying brain function, brain diseases, and large-scale brain models or connectomes. We review the driving forces behind big data and HPC methods applied to brain science, including deep learning, powerful data analysis capabilities, and computational performance solutions, each of which can be used to improve diagnostic accuracy and research output. This work reinforces predictions that big data and HPC will continue to improve brain science by making ultrahigh-performance analysis possible, by improving data standardization and sharing, and by providing new neuromorphic insights.

Machine learning applications to clinical decision support in neurosurgery: an artificial intelligence augmented systematic review

Machine learning (ML) involves algorithms learning patterns in large, complex datasets to predict and classify. Algorithms include neural networks (NN), logistic regression (LR), and support vector machines (SVM). ML may generate substantial improvements in neurosurgery. This systematic review assessed the current state of neurosurgical ML applications and the performance of algorithms applied. Our systematic search strategy yielded 6866 results, 70 of which met inclusion criteria. Performance statistics analyzed included area under the receiver operating characteristics curve (AUC), accuracy, sensitivity, and specificity. Natural language processing (NLP) was used to model topics across the corpus and to identify keywords within surgical subspecialties. ML applications were heterogeneous. The densest cluster of studies focused on preoperative evaluation, planning, and outcome prediction in spine surgery. The main algorithms applied were NN, LR, and SVM. Input and output features varied widely and were listed to facilitate future research. The accuracy (F_(2,19) = 6.56, p < 0.01) and specificity (F_(2,16) = 5.57, p < 0.01) of NN, LR, and SVM differed significantly. NN algorithms demonstrated significantly higher accuracy than LR. SVM demonstrated significantly higher specificity than LR. We found no significant difference between NN, LR, and SVM AUC and sensitivity. NLP topic modeling reached maximum coherence at seven topics, which were defined by modeling approach, surgery type, and pathology themes. Keywords captured research foci within surgical domains. ML technology accurately predicts outcomes and facilitates clinical decision-making in neurosurgery. NNs frequently outperformed other algorithms on supervised learning tasks. This study identified gaps in the literature and opportunities for future neurosurgical ML research.

Single-Cell Recordings to Target the Anterior Nucleus of the Thalamus in Deep Brain Stimulation for Patients with Refractory Epilepsy

Deep brain stimulation (DBS) of the anterior nucleus of the thalamus (ANT) is a promising treatment for patients with refractory epilepsy. However, therapy response varies and precise positioning of the DBS lead is potentially essential for maximizing therapeutic efficacy. We investigate if single-cell recordings acquired by microelectrode recordings can aid targeting of the ANT during surgery and hypothesize that the neuronal firing properties of the target region relate to clinical outcome. We prospectively included 10 refractory epilepsy patients and performed microelectrode recordings under general anesthesia to identify the change in neuronal signals when approaching and transecting the ANT. The neuronal firing properties of the target region, anatomical locations of microelectrode recordings and active contact positions of the DBS lead along the recorded trajectory were compared between responders and nonresponders to DBS. We obtained 19 sets of recordings from 10 patients (five responders and five nonresponders). Amongst the 403 neurons detected, 365 (90.6%) were classified as bursty. Entry into the ANT was characterized by an increase in firing rate while exit of the ANT was characterized by a decrease in firing rate. Comparing the trajectories of responders to nonresponders, we found differences neither in the neuronal firing properties themselves nor in their locations relative to the position of the active contact. Single-cell firing rate acquired by microelectrode recordings under general anesthesia can thus aid targeting of the ANT during surgery, but is not related to clinical outcome in DBS for patients with refractory epilepsy.

Finding the balance between model complexity and performance: Using ventral striatal oscillations to classify feeding behavior in rats

The ventral striatum (VS) is a central node within a distributed network that controls appetitive behavior, and neuromodulation of the VS has demonstrated therapeutic potential for appetitive disorders. Local field potential (LFP) oscillations recorded from deep brain stimulation (DBS) electrodes within the VS are a pragmatic source of neural systems-level information about appetitive behavior that could be used in responsive neuromodulation systems. Here, we recorded LFPs from the bilateral nucleus accumbens core and shell (subregions of the VS) during limited access to palatable food across varying conditions of hunger and food palatability in male rats. We used standard statistical methods (logistic regression) as well as the machine learning algorithm lasso to predict aspects of feeding behavior using VS LFPs. We were able to predict the amount of food eaten, the increase in consumption following food deprivation, and the type of food eaten. Further, we were able to predict whether the initiation of feeding was imminent up to 42.5 seconds before feeding began and classify current behavior as either feeding or not-feeding. In classifying feeding behavior, we found an optimal balance between model complexity and performance with models using 3 LFP features primarily from the alpha and high gamma frequencies. As shown here, unbiased methods can identify systems-level neural activity linked to domains of mental illness with potential application to the development and personalization of novel treatments.

As neuropsychiatry begins to leverage the power of computational methods to understand disease states and to develop better therapies, it is vital that we acknowledge the trade-offs between model complexity and performance. We show that computational methods can elucidate a neural signature of feeding behavior and we show how these methods could be used to discover neural patterns related to other behaviors and reveal new potential therapeutic targets. Further, our results help to contextualize both the limitations and potential of applying computational methods to neuropsychiatry by showing how changing the data being used to train predictive models (e.g., population vs. individual data) can have a large impact on how model performance generalizes across time, internal states, and individuals.

Stimulation of the Tractography-Defined Subthalamic Nucleus Regions Correlates With Clinical Outcomes

BACKGROUND

Although deep brain stimulation (DBS) of the dorsolateral subthalamic nucleus (STN) is a well-established surgical treatment for patients with Parkinson disease (PD), there is still controversy about the relationship between the functional segregation of the STN and clinical outcomes.

OBJECTIVE

To correlate motor and neuropsychological (NPS) outcomes with the overlap between the volume of activated tissue (VAT) and the tractography-defined regions within the STN.

METHODS

Retrospective study evaluating 13 patients with PD treated with STN-DBS. With the aid of tractography, the STN was segmented into 4 regions: smaSTN (supplementary motor area STN), m1STN (primary motor area STN), mSTN (the sum of the m1STN and the smaSTN segments), and nmSTN (non-motor STN). We computed the overlap coefficients between these STN regions and the patient-specific VAT. The VAT outside of the STN was also calculated. These coefficients were then correlated with motor (Unified Parkinson's Disease Rating Scale, UPDRS III) and NPS outcomes.

RESULTS

Stimulation of the mSTN segment was significantly correlated with UPDRS III and bradykinesia improvement. Stimulation of the smaSTN segment, but not the m1STN one, had a positive correlation with bradykinesia improvement. Stimulation of the nmSTN segment was negatively correlated with the improvement in rigidity. Stimulation outside of the STN was correlated with some beneficial NPS effects.

CONCLUSION

Stimulation of the tractography-defined motor STN, mainly the smaSTN segment, is positively correlated with motor outcomes, whereas stimulation of the nmSTN is correlated with poor motor outcomes. Further validation of these results might help individualize and optimize targets prior to STN-DBS.

Systems approaches to optimizing deep brain stimulation therapies in Parkinson's disease

Over the last 30 years, deep brain stimulation (DBS) has been used to treat chronic neurological diseases like dystonia, obsessive-compulsive disorders, essential tremor, Parkinson's disease, and more recently, dementias, depression, cognitive disorders, and epilepsy. Despite its wide use, DBS presents numerous challenges for both clinicians and engineers. One challenge is the design of novel, more efficient DBS therapies, which are hampered by the lack of complete understanding about the cellular mechanisms of therapeutic DBS. Another challenge is the existence of redundancy in clinical outcomes, that is, different DBS programs can result in similar clinical benefits but very little information (e.g., predictive models, longitudinal data, metrics, etc.) is available to select one program over another. Finally, there is high variability in patients' responses to DBS, which forces clinicians to carefully adjust the stimulation settings to each patient via lengthy programming sessions. Researchers in neural engineering and systems biology have been tackling these challenges over the past few years with the specific goal of developing novel DBS therapies, design methodologies, and computational tools that optimize the therapeutic effects of DBS in each patient. Furthermore, efforts are being made to automatically adapt the DBS treatment to the fluctuations of disease symptoms. A review of the quantitative approaches currently available for the treatment of Parkinson's disease is presented here with an emphasis on the contributions that systems theoretical approaches have provided to understand the global dynamics of complex neuronal circuits in the brain under DBS. This article is categorized under: Translational, Genomic, and Systems Medicine > Therapeutic Methods Analytical and Computational Methods > Computational Methods Analytical and Computational Methods > Dynamical Methods Physiology > Mammalian Physiology in Health and Disease.

Patient-specific anatomical model for deep brain stimulation based on 7 Tesla MRI

Objective

Deep brain stimulation (DBS) requires accurate localization of the anatomical target structure, and the precise placement of the DBS electrode within it. Ultra-high field 7 Tesla (T) MR images can be utilized to create patient-specific anatomical 3D models of the subthalamic nuclei (STN) to enhance pre-surgical DBS targeting as well as post-surgical visualization of the DBS lead position and orientation. We validated the accuracy of the 7T imaging-based patient-specific model of the STN and measured the variability of the location and dimensions across movement disorder patients.

Methods

72 patients who underwent DBS surgery were scanned preoperatively on 7T MRI. Segmentations and 3D volume rendering of the STN were generated for all patients. For 21 STN-DBS cases, microelectrode recording (MER) was used to validate the segmentation. For 12 cases, we computed the correlation between the overlap of the STN and volume of tissue activated (VTA) and the monopolar review for a further validation of the model’s accuracy and its clinical relevancy.

Results

We successfully reconstructed and visualized the STN in all patients. Significant variability was found across individuals regarding the location of the STN center of mass as well as its volume, length, depth and width. Significant correlations were found between MER and the 7T imaging-based model of the STN (r = 0.86) and VTA-STN overlap and the monopolar review outcome (r = 0.61).

Conclusion

The results suggest that an accurate visualization and localization of a patient-specific 3D model of the STN can be generated based on 7T MRI. The imaging-based 7T MRI STN model was validated using MER and patient’s clinical outcomes. The significant variability observed in the STN location and shape based on a large number of patients emphasizes the importance of an accurate direct visualization of the STN for DBS targeting. An accurate STN localization can facilitate postoperative stimulation parameters for optimized patient outcome.

Surface EEG-Transcranial Direct Current Stimulation (tDCS) Closed-Loop System

Conventional transcranial direct current stimulation (tDCS) protocols rely on applying electrical current at a fixed intensity and duration without using surrogate markers to direct the interventions. This has led to some mixed results; especially because tDCS induced effects may vary depending on the ongoing level of brain activity. Therefore, the objective of this preliminary study was to assess the feasibility of an EEG-triggered tDCS system based on EEG online analysis of its frequency bands. Six healthy volunteers were randomized to participate in a double-blind sham-controlled crossover design to receive a single session of 10[Formula: see text]min 2[Formula: see text]mA cathodal and sham tDCS. tDCS trigger controller was based upon an algorithm designed to detect an increase in the relative beta power of more than 200%, accompanied by a decrease of 50% or more in the relative alpha power, based on baseline EEG recordings. EEG-tDCS closed-loop-system was able to detect the predefined EEG magnitude deviation and successfully triggered the stimulation in all participants. This preliminary study represents a proof-of-concept for the development of an EEG-tDCS closed-loop system in humans. We discuss and review here different methods of closed loop system that can be considered and potential clinical applications of such system.

Computer-Guided Deep Brain Stimulation Programming for Parkinson's Disease

OBJECTIVE

Pilot study to evaluate computer-guided deep brain stimulation (DBS) programming designed to optimize stimulation settings using objective motion sensor-based motor assessments.

MATERIALS AND METHODS

Seven subjects (five males; 54-71 years) with Parkinson's disease (PD) and recently implanted DBS systems participated in this pilot study. Within two months of lead implantation, the subject returned to the clinic to undergo computer-guided programming and parameter selection. A motion sensor was placed on the index finger of the more affected hand. Software guided a monopolar survey during which monopolar stimulation on each contact was iteratively increased followed by an automated assessment of tremor and bradykinesia. After completing assessments at each setting, a software algorithm determined stimulation settings designed to minimize symptom severities, side effects, and battery usage.

RESULTS

Optimal DBS settings were chosen based on average severity of motor symptoms measured by the motion sensor. Settings chosen by the software algorithm identified a therapeutic window and improved tremor and bradykinesia by an average of 35.7% compared with baseline in the "off" state (p < 0.01).

CONCLUSIONS

Motion sensor-based computer-guided DBS programming identified stimulation parameters that significantly improved tremor and bradykinesia with minimal clinician involvement. Automated motion sensor-based mapping is worthy of further investigation and may one day serve to extend programming to populations without access to specialized DBS centers.

Modeling sequence and quasi-uniform assumption in computational neurostimulation

Computational neurostimulation aims to develop mathematical constructs that link the application of neuromodulation with changes in behavior and cognition. This process is critical but daunting for technical challenges and scientific unknowns. The overarching goal of this review is to address how this complex task can be made tractable. We describe a framework of sequential modeling steps to achieve this: (1) current flow models, (2) cell polarization models, (3) network and information processing models, and (4) models of the neuroscientific correlates of behavior. Each step is explained with a specific emphasis on the assumptions underpinning underlying sequential implementation. We explain the further implementation of the quasi-uniform assumption to overcome technical limitations and unknowns. We specifically focus on examples in electrical stimulation, such as transcranial direct current stimulation. Our approach and conclusions are broadly applied to immediate and ongoing efforts to deploy computational neurostimulation.