Reaching the ceiling? Empirical scaling behaviour for deep EEG pathology classification.

Machine learning techniques, particularly deep convolutional neural networks (ConvNets), are increasingly being used to automate clinical EEG analysis, with the potential to reduce the clinical burden and improve patient care. However, further research is required before they can be used in clinical settings, particularly regarding the impact of the number of training samples and model parameters on their testing error. To address this, we present a comprehensive study of the empirical scaling behaviour of ConvNets for EEG pathology classification. We analysed the testing error with increasing the training samples and model size for four different ConvNet architectures. The focus of our experiments is width scaling, and we have increased the number of parameters to up to 1.8 million. Our evaluation was based on two publicly available datasets: the Temple University Hospital (TUH) Abnormal EEG Corpus and the TUH Abnormal Expansion Balanced EEG Corpus, which together contain 10,707 training samples. The results show that the testing error follows a saturating power-law with both model and dataset size. This pattern is consistent across different datasets and ConvNet architectures. Furthermore, empirically observed accuracies saturate at 85%-87%, which may be due to an imperfect inter-rater agreement on the clinical labels. The empirical scaling behaviour of the test performance with dataset and model size has significant implications for deep EEG pathology classification research and practice. Our findings highlight the potential of deep ConvNets for high-performance EEG pathology classification, and the identified scaling relationships provide valuable recommendations for the advancement of automated EEG diagnostics.

An extended clinical EEG dataset with 15,300 automatically labelled recordings for pathology decoding.

Automated clinical EEG analysis using machine learning (ML) methods is a growing EEG research area. Previous studies on binary EEG pathology decoding have mainly used the Temple University Hospital (TUH) Abnormal EEG Corpus (TUAB) which contains approximately 3,000 manually labelled EEG recordings. To evaluate and eventually even improve the generalisation performance of machine learning methods for EEG pathology, decoding larger, publicly available datasets is required. A number of studies addressed the automatic labelling of large open-source datasets as an approach to create new datasets for EEG pathology decoding, but little is known about the extent to which training on larger, automatically labelled dataset affects decoding performances of established deep neural networks. In this study, we automatically created additional pathology labels for the Temple University Hospital (TUH) EEG Corpus (TUEG) based on the medical reports using a rule-based text classifier. We generated a dataset of 15,300 newly labelled recordings, which we call the TUH Abnormal Expansion EEG Corpus (TUABEX), and which is five times larger than the TUAB. Since the TUABEX contains more pathological (75%) than non-pathological (25%) recordings, we then selected a balanced subset of 8,879 recordings, the TUH Abnormal Expansion Balanced EEG Corpus (TUABEXB). To investigate how training on a larger, automatically labelled dataset affects the decoding performance of deep neural networks, we applied four established deep convolutional neural networks (ConvNets) to the task of pathological versus non-pathological classification and compared the performance of each architecture after training on different datasets. The results show that training on the automatically labelled TUABEXB dataset rather than training on the manually labelled TUAB dataset increases accuracies on TUABEXB and even for TUAB itself for some architectures. We argue that automatically labelling of large open-source datasets can be used to efficiently utilise the massive amount of EEG data stored in clinical archives. We make the proposed TUABEXB available open source and thus offer a new dataset for EEG machine learning research.

Patient groups in Rheumatoid arthritis identified by deep learning respond differently to biologic or targeted synthetic DMARDs.

Cycling of biologic or targeted synthetic disease modifying antirheumatic drugs (b/tsDMARDs) in rheumatoid arthritis (RA) patients due to non-response is a problem preventing and delaying disease control. We aimed to assess and validate treatment response of b/tsDMARDs among clusters of RA patients identified by deep learning. We clustered RA patients clusters at first-time b/tsDMARD (cohort entry) in the Swiss Clinical Quality Management in Rheumatic Diseases registry (SCQM) [1999-2018]. We performed comparative effectiveness analyses of b/tsDMARDs (ref. adalimumab) using Cox proportional hazard regression. Within 15 months, we assessed b/tsDMARD stop due to non-response, and separately a ≥20% reduction in DAS28-esr as a response proxy. We validated results through stratified analyses according to most distinctive patient characteristics of clusters. Clusters comprised between 362 and 1481 patients (3516 unique patients). Stratified (validation) analyses confirmed comparative effectiveness results among clusters: Patients with ≥2 conventional synthetic DMARDs and prednisone at b/tsDMARD initiation, male patients, as well as patients with a lower disease burden responded better to tocilizumab than to adalimumab (hazard ratio [HR] 5.46, 95% confidence interval [CI] [1.76-16.94], and HR 8.44 [3.43-20.74], and HR 3.64 [2.04-6.49], respectively). Furthermore, seronegative women without use of prednisone at b/tsDMARD initiation as well as seropositive women with a higher disease burden and longer disease duration had a higher risk of non-response with golimumab (HR 2.36 [1.03-5.40] and HR 5.27 [2.10-13.21], respectively) than with adalimumab. Our results suggest that RA patient clusters identified by deep learning may have different responses to first-line b/tsDMARD. Thus, it may suggest optimal first-line b/tsDMARD for certain RA patients, which is a step forward towards personalizing treatment. However, further research in other cohorts is needed to verify our results.

Personalized prediction of disease activity in patients with rheumatoid arthritis using an adaptive deep neural network.

BACKGROUND: Deep neural networks learn from former experiences on a large scale and can be used to predict future disease activity as potential clinical decision support. AdaptiveNet is a novel adaptive recurrent neural network optimized to deal with heterogeneous and missing clinical data. OBJECTIVE: We investigate AdaptiveNet for the prediction of individual disease activity in patients from a rheumatoid arthritis (RA) registry. METHODS: Demographic and disease characteristics from over 9500 patients and 65.000 visits from the Swiss Quality Management (SCQM) database were used to train and evaluate the network. Patient characteristics, clinical and patient reported outcomes, laboratory values and medication were used as input features. DAS28-BSR served as a target to predict active RA and future numeric individual disease activity by classification and regression. RESULTS: AdaptiveNet predicted active disease defined as DAS28-BSR >2.6 at the next visit with an overall accuracy of 75.6% (SD +- 0.7%) and a sensitivity and specificity of 84.2% (SD +- 1.6%) and 61.5% (SD +- 3.6%), respectively. Prediction performance was significantly higher in patients with a disease duration >3 years and positive rheumatoid factor. Regression allowed forecasting individual DAS28-BSR values with a mean squared error (MSE) of 0.9 (SD +- 0.05). This corresponds to a 8% deviation between estimated and real DAS28-BSR values. Compared to linear regression, random forest and support vector machines, AdaptiveNet showed an increased performance of over 7% in MSE. Medication played a minor role in the prediction of RA disease activity. CONCLUSION: AdaptiveNet has a superior capacity to predict numeric RA disease activity compared to classical machine learning architectures. All investigated models had limitations in low specificity.

Machine-learning-based diagnostics of EEG pathology.

Machine learning (ML) methods have the potential to automate clinical EEG analysis. They can be categorized into feature-based (with handcrafted features), and end-to-end approaches (with learned features). Previous studies on EEG pathology decoding have typically analyzed a limited number of features, decoders, or both. For a I) more elaborate feature-based EEG analysis, and II) in-depth comparisons of both approaches, here we first develop a comprehensive feature-based framework, and then compare this framework to state-of-the-art end-to-end methods. To this aim, we apply the proposed feature-based framework and deep neural networks including an EEG-optimized temporal convolutional network (TCN) to the task of pathological versus non-pathological EEG classification. For a robust comparison, we chose the Temple University Hospital (TUH) Abnormal EEG Corpus (v2.0.0), which contains approximately 3000 EEG recordings. The results demonstrate that the proposed feature-based decoding framework can achieve accuracies on the same level as state-of-the-art deep neural networks. We find accuracies across both approaches in an astonishingly narrow range from 81 to 86%. Moreover, visualizations and analyses indicated that both approaches used similar aspects of the data, e.g., delta and theta band power at temporal electrode locations. We argue that the accuracies of current binary EEG pathology decoders could saturate near 90% due to the imperfect inter-rater agreement of the clinical labels, and that such decoders are already clinically useful, such as in areas where clinical EEG experts are rare. We make the proposed feature-based framework available open source and thus offer a new tool for EEG machine learning research.

Applied machine learning and artificial intelligence in rheumatology.

Machine learning as a field of artificial intelligence is increasingly applied in medicine to assist patients and physicians. Growing datasets provide a sound basis with which to apply machine learning methods that learn from previous experiences. This review explains the basics of machine learning and its subfields of supervised learning, unsupervised learning, reinforcement learning and deep learning. We provide an overview of current machine learning applications in rheumatology, mainly supervised learning methods for e-diagnosis, disease detection and medical image analysis. In the future, machine learning will be likely to assist rheumatologists in predicting the course of the disease and identifying important disease factors. Even more interestingly, machine learning will probably be able to make treatment propositions and estimate their expected benefit (e.g. by reinforcement learning). Thus, in future, shared decision-making will not only include the patient's opinion and the rheumatologist's empirical and evidence-based experience, but it will also be influenced by machine-learned evidence.

Hybrid Brain-Computer-Interfacing for Human-Compliant Robots: Inferring Continuous Subjective Ratings With Deep Regression.

Appropriate robot behavior during human-robot interaction is a key part in the development of human-compliant assistive robotic systems. This study poses the question of how to continuously evaluate the quality of robotic behavior in a hybrid brain-computer interfacing (BCI) task, combining brain and non-brain signals, and how to use the collected information to adapt the robot's behavior accordingly. To this aim, we developed a rating system compatible with EEG recordings, requiring the users to execute only small movements with their thumb on a wireless controller to rate the robot's behavior on a continuous scale. The ratings were recorded together with dry EEG, respiration, ECG, and robotic joint angles in ROS. Pilot experiments were conducted with three users that had different levels of previous experience with robots. The results demonstrate the feasibility to obtain continuous rating data that give insight into the subjective user perception during direct human-robot interaction. The rating data suggests differences in subjective perception for users with no, moderate, or substantial previous robot experience. Furthermore, a variety of regression techniques, including deep CNNs, allowed us to predict the subjective ratings. Performance was better when using the position of the robotic hand than when using EEG, ECG, or respiration. A consistent advantage of features expected to be related to a motor bias could not be found. Across-user predictions showed that the models most likely learned a combination of general and individual features across-users. A transfer of pre-trained regressor to a new user was especially accurate in users with more experience. For future research, studies with more participants will be needed to evaluate the methodology for its use in practice. Data and code to reproduce this study are available at https://github.com/TNTLFreiburg/NiceBot.

Hardware Implementation of a Performance and Energy-optimized Convolutional Neural Network for Seizure Detection.

We present for the first time a µW-power convolutional neural network for seizure detection running on a low-power microcontroller. On a dataset of 22 patients a median sensitivity of 100% is achieved. With a false positive rate of 20.7 fp/h and a short detection delay of 3.4 s it is suitable for the application in an implantable closed-loop device.

Autonomous Optimization of Targeted Stimulation of Neuronal Networks.

Driven by clinical needs and progress in neurotechnology, targeted interaction with neuronal networks is of increasing importance. Yet, the dynamics of interaction between intrinsic ongoing activity in neuronal networks and their response to stimulation is unknown. Nonetheless, electrical stimulation of the brain is increasingly explored as a therapeutic strategy and as a means to artificially inject information into neural circuits. Strategies using regular or event-triggered fixed stimuli discount the influence of ongoing neuronal activity on the stimulation outcome and are therefore not optimal to induce specific responses reliably. Yet, without suitable mechanistic models, it is hardly possible to optimize such interactions, in particular when desired response features are network-dependent and are initially unknown. In this proof-of-principle study, we present an experimental paradigm using reinforcement-learning (RL) to optimize stimulus settings autonomously and evaluate the learned control strategy using phenomenological models. We asked how to (1) capture the interaction of ongoing network activity, electrical stimulation and evoked responses in a quantifiable 'state' to formulate a well-posed control problem, (2) find the optimal state for stimulation, and (3) evaluate the quality of the solution found. Electrical stimulation of generic neuronal networks grown from rat cortical tissue in vitro evoked bursts of action potentials (responses). We show that the dynamic interplay of their magnitudes and the probability to be intercepted by spontaneous events defines a trade-off scenario with a network-specific unique optimal latency maximizing stimulus efficacy. An RL controller was set to find this optimum autonomously. Across networks, stimulation efficacy increased in 90% of the sessions after learning and learned latencies strongly agreed with those predicted from open-loop experiments. Our results show that autonomous techniques can exploit quantitative relationships underlying activity-response interaction in biological neuronal networks to choose optimal actions. Simple phenomenological models can be useful to validate the quality of the resulting controllers.

Modeling effects of intrinsic and extrinsic rewards on the competition between striatal learning systems.

A common assumption in psychology, economics, and other fields holds that higher performance will result if extrinsic rewards (such as money) are offered as an incentive. While this principle seems to work well for tasks that require the execution of the same sequence of steps over and over, with little uncertainty about the process, in other cases, especially where creative problem solving is required due to the difficulty in finding the optimal sequence of actions, external rewards can actually be detrimental to task performance. Furthermore, they have the potential to undermine intrinsic motivation to do an otherwise interesting activity. In this work, we extend a computational model of the dorsomedial and dorsolateral striatal reinforcement learning systems to account for the effects of extrinsic and intrinsic rewards. The model assumes that the brain employs both a goal-directed and a habitual learning system, and competition between both is based on the trade-off between the cost of the reasoning process and value of information. The goal-directed system elicits internal rewards when its models of the environment improve, while the habitual system, being model-free, does not. Our results account for the phenomena that initial extrinsic reward leads to reduced activity after extinction compared to the case without any initial extrinsic rewards, and that performance in complex task settings drops when higher external rewards are promised. We also test the hypothesis that external rewards bias the competition in favor of the computationally efficient, but cruder and less flexible habitual system, which can negatively influence intrinsic motivation and task performance in the class of tasks we consider.

Information processing in echo state networks at the edge of chaos.

We investigate information processing in randomly connected recurrent neural networks. It has been shown previously that the computational capabilities of these networks are maximized when the recurrent layer is close to the border between a stable and an unstable dynamics regime, the so called edge of chaos. The reasons, however, for this maximized performance are not completely understood. We adopt an information-theoretical framework and are for the first time able to quantify the computational capabilities between elements of these networks directly as they undergo the phase transition to chaos. Specifically, we present evidence that both information transfer and storage in the recurrent layer are maximized close to this phase transition, providing an explanation for why guiding the recurrent layer toward the edge of chaos is computationally useful. As a consequence, our study suggests self-organized ways of improving performance in recurrent neural networks, driven by input data. Moreover, the networks we study share important features with biological systems such as feedback connections and online computation on input streams. A key example is the cerebral cortex, which was shown to also operate close to the edge of chaos. Consequently, the behavior of model systems as studied here is likely to shed light on reasons why biological systems are tuned into this specific regime.

Initialization and self-organized optimization of recurrent neural network connectivity.

Reservoir computing (RC) is a recent paradigm in the field of recurrent neural networks. Networks in RC have a sparsely and randomly connected fixed hidden layer, and only output connections are trained. RC networks have recently received increased attention as a mathematical model for generic neural microcircuits to investigate and explain computations in neocortical columns. Applied to specific tasks, their fixed random connectivity, however, leads to significant variation in performance. Few problem-specific optimization procedures are known, which would be important for engineering applications, but also in order to understand how networks in biology are shaped to be optimally adapted to requirements of their environment. We study a general network initialization method using permutation matrices and derive a new unsupervised learning rule based on intrinsic plasticity (IP). The IP-based learning uses only local learning, and its aim is to improve network performance in a self-organized way. Using three different benchmarks, we show that networks with permutation matrices for the reservoir connectivity have much more persistent memory than the other methods but are also able to perform highly nonlinear mappings. We also show that IP-based on sigmoid transfer functions is limited concerning the output distributions that can be achieved.