A Shared Vision for Machine Learning in Neuroscience

Graph-informed convolutional autoencoder to classify brain responses during sleep

Automated machine-learning algorithms that analyze biomedical signals have been used to identify sleep patterns and health issues. However, their performance is often suboptimal, especially when dealing with imbalanced datasets. In this paper, we present a robust sleep state (SlS) classification algorithm utilizing electroencephalogram (EEG) signals. To this aim, we pre-processed EEG recordings from 33 healthy subjects. Then, functional connectivity features and recurrence quantification analysis were extracted from sub-bands. The graphical representation was calculated from phase locking value, coherence, and phase-amplitude coupling. Statistical analysis was used to select features with p-values of less than 0.05. These features were compared between four states: wakefulness, non-rapid eye movement (NREM) sleep, rapid eye movement (REM) sleep during presenting auditory stimuli, and REM sleep without stimuli. Eighteen types of different stimuli including instrumental and natural sounds were presented to participants during REM. The selected significant features were used to train a novel deep-learning classifiers. We designed a graph-informed convolutional autoencoder called GICA to extract high-level features from the functional connectivity features. Furthermore, an attention layer based on recurrence rate features extracted from EEGs was incorporated into the GICA classifier to enhance the dynamic ability of the model. The proposed model was assessed by comparing it to baseline systems in the literature. The accuracy of the SlS-GICA classifier is 99.92% on the significant feature set. This achievement could be considered in real-time and automatic applications to develop new therapeutic strategies for sleep-related disorders.

The NERVE-ML (neural engineering reproducibility and validity essentials for machine learning) checklist: ensuring machine learning advances neural engineering. J Neural Eng 2025;22:021002. [PMID: 40073450 PMCID: PMC11948487 DOI: 10.1088/1741-2552/adbfbd] [Citation(s) in RCA: 0] [Impact Index Per Article: 0]

Objective.Machine learning's (MLs) ability to capture intricate patterns makes it vital in neural engineering research. With its increasing use, ensuring the validity and reproducibility of ML methods is critical. Unfortunately, this has not always been the case in practice, as there have been recent retractions across various scientific fields due to the misuse of ML methods and validation procedures. To address these concerns, we propose the first version of the neural engineering reproducibility and validity essentials for ML (NERVE-ML) checklist, a framework designed to promote the transparent, reproducible, and valid application of ML in neural engineering.Approach.We highlight some of the unique challenges of model validation in neural engineering, including the difficulties from limited subject numbers, repeated or non-independent samples, and high subject heterogeneity. Through detailed case studies, we demonstrate how different validation approaches can lead to divergent scientific conclusions, highlighting the importance of selecting appropriate procedures guided by the NERVE-ML checklist. Effectively addressing these challenges and properly scoping scientific conclusions will ensure that ML contributes to, rather than hinders, progress in neural engineering.Main results.Our case studies demonstrate that improper validation approaches can result in flawed studies or overclaimed scientific conclusions, complicating the scientific discourse. The NERVE-ML checklist effectively addresses these concerns by providing guidelines to ensure that ML approaches in neural engineering are reproducible and lead to valid scientific conclusions.Significance.By effectively addressing these challenges and properly scoping scientific conclusions guided by the NERVE-ML checklist, we aim to help pave the way for a future where ML reliably enhances the quality and impact of neural engineering research.

Depression diagnosis: EEG-based cognitive biomarkers and machine learning

Depression is a complex mental illness that has significant effects on people as well as society. The traditional techniques for the diagnosis of depression, along with the potential of nascent biomarkers especially EEG-based biomarkers, are studied. This review explores the significance of cognitive biomarkers, offering a comprehensive understanding of their role in the overall assessment of depression. It also investigates the effects of depression on various brain regions, outlines promising areas for future research, and emphasizes the importance of understanding the neurophysiological roots of depression. Furthermore, it elucidates how machine learning and deep learning models are integrated into EEG-based depression diagnosis, emphasizing their importance in optimizing personalized therapeutic protocols and improving diagnostic accuracy with EEG data analysis.

Eeg Microstates and Balance Parameters for Stroke Discrimination: A Machine Learning Approach

Electroencephalography microstates (EEG-MS) show promise to be a neurobiological biomarker in stroke. Thus, the aim of the study was to identify biomarkers to discriminate stroke patients from healthy individuals based on EEG-MS and clinical features using a machine learning approach. Fifty-four participants (27 stroke patients and 27 healthy age and sex-matched controls) were recruited. We recorded EEG-MS using 32 channels during eyes-closed and eyes-open conditions and analyzed the four classical EEG-MS maps (A, B, C, D). Clinical information and motor aspects were evaluated. A machine learning method using k-means algorithms to discriminate stroke patients from healthy subjects showed that the most influential parameters in clustering were balance scores and microstate parameters (duration and coverage of microstate A, duration, coverage and occurrence of microstates C and global variance explained). To evaluate the quality of clustering, the Silhouette score was applied and the score was close to 0.20, indicating that the clusters overlap. These results are encouraging and support the usefulness of these methods for classifying stroke patients in order to contribute to the development of therapeutic strategies, improve the clinical management of these patients, and consequently reduce the associated costs.

The application of eXplainable artificial intelligence in studying cognition: A scoping review

The rapid advancement of artificial intelligence (AI) has sparked renewed discussions on its trustworthiness and the concept of eXplainable AI (XAI). Recent research in neuroscience has emphasized the relevance of XAI in studying cognition. This scoping review aims to identify and analyze various XAI methods used to study the mechanisms and features of cognitive function and dysfunction. In this study, the collected evidence is qualitatively assessed to develop an effective framework for approaching XAI in cognitive neuroscience. Based on the Joanna Briggs Institute and preferred reporting items for systematic reviews and meta-analyses extension for scoping review guidelines, we searched for peer-reviewed articles on MEDLINE, Embase, Web of Science, Cochrane Central Register of Controlled Trials, and Google Scholar. Two reviewers performed data screening, extraction, and thematic analysis in parallel. Twelve eligible experimental studies published in the past decade were included. The results showed that the majority (75%) focused on normal cognitive functions such as perception, social cognition, language, executive function, and memory, while others (25%) examined impaired cognition. The predominant XAI methods employed were intrinsic XAI (58.3%), followed by attribution-based (41.7%) and example-based (8.3%) post hoc methods. Explainability was applied at a local (66.7%) or global (33.3%) scope. The findings, predominantly correlational, were anatomical (83.3%) or nonanatomical (16.7%). In conclusion, while these XAI techniques were lauded for their predictive power, robustness, testability, and plausibility, limitations included oversimplification, confounding factors, and inconsistencies. The reviewed studies showcased the potential of XAI models while acknowledging current challenges in causality and oversimplification, particularly emphasizing the need for reproducibility.

Simple Behavioral Analysis (SimBA) as a platform for explainable machine learning in behavioral neuroscience

The study of complex behaviors is often challenging when using manual annotation due to the absence of quantifiable behavioral definitions and the subjective nature of behavioral annotation. Integration of supervised machine learning approaches mitigates some of these issues through the inclusion of accessible and explainable model interpretation. To decrease barriers to access, and with an emphasis on accessible model explainability, we developed the open-source Simple Behavioral Analysis (SimBA) platform for behavioral neuroscientists. SimBA introduces several machine learning interpretability tools, including SHapley Additive exPlanation (SHAP) scores, that aid in creating explainable and transparent behavioral classifiers. Here we show how the addition of explainability metrics allows for quantifiable comparisons of aggressive social behavior across research groups and species, reconceptualizing behavior as a sharable reagent and providing an open-source framework. We provide an open-source, graphical user interface (GUI)-driven, well-documented package to facilitate the movement toward improved automation and sharing of behavioral classification tools across laboratories.

Narcissus reflected: Grey and white matter features joint contribution to the default mode network in predicting narcissistic personality traits

Despite the clinical significance of narcissistic personality, its neural bases have not been clarified yet, primarily because of methodological limitations of the previous studies, such as the low sample size, the use of univariate techniques and the focus on only one brain modality. In this study, we employed for the first time a combination of unsupervised and supervised machine learning methods, to identify the joint contributions of grey matter (GM) and white matter (WM) to narcissistic personality traits (NPT). After preprocessing, the brain scans of 135 participants were decomposed into eight independent networks of covarying GM and WM via parallel ICA. Subsequently, stepwise regression and Random Forest were used to predict NPT. We hypothesized that a fronto-temporo parietal network, mainly related to the default mode network, may be involved in NPT and associated WM regions. Results demonstrated a distributed network that included GM alterations in fronto-temporal regions, the insula and the cingulate cortex, along with WM alterations in cerebellar and thalamic regions. To assess the specificity of our findings, we also examined whether the brain network predicting narcissism could also predict other personality traits (i.e., histrionic, paranoid and avoidant personalities). Notably, this network did not predict such personality traits. Additionally, a supervised machine learning model (Random Forest) was used to extract a predictive model for generalization to new cases. Results confirmed that the same network could predict new cases. These findings hold promise for advancing our understanding of personality traits and potentially uncovering brain biomarkers associated with narcissism.

An unbiased method to partition diverse neuronal responses into functional ensembles reveals interpretable population dynamics during innate social behavior

In neuroscience, understanding how single-neuron firing contributes to distributed neural ensembles is crucial. Traditional methods of analysis have been limited to descriptions of whole population activity, or, when analyzing individual neurons, criteria for response categorization varied significantly across experiments. Current methods lack scalability for large datasets, fail to capture temporal changes and rely on parametric assumptions. There's a need for a robust, scalable, and non-parametric functional clustering approach to capture interpretable dynamics. To address this challenge, we developed a model-based, statistical framework for unsupervised clustering of multiple time series datasets that exhibit nonlinear dynamics into an a-priori-unknown number of parameterized ensembles called Functional Encoding Units (FEUs). FEU outperforms existing techniques in accuracy and benchmark scores. Here, we apply this FEU formalism to single-unit recordings collected during social behaviors in rodents and primates and demonstrate its hypothesis-generating and testing capacities. This novel pipeline serves as an analytic bridge, translating neural ensemble codes across model systems.

Predicting interindividual response to theta burst stimulation in the lower limb motor cortex using machine learning

Using theta burst stimulation (TBS) to induce neural plasticity has played an important role in improving the treatment of neurological disorders. However, the variability of TBS-induced synaptic plasticity in the primary motor cortex prevents its clinical application. Thus, factors associated with this variability should be explored to enable the creation of a predictive model. Statistical approaches, such as regression analysis, have been used to predict the effects of TBS. Machine learning may potentially uncover previously unexplored predictive factors due to its increased capacity for capturing nonlinear changes. In this study, we used our prior dataset (Katagiri et al., 2020) to determine the factors that predict variability in TBS-induced synaptic plasticity in the lower limb motor cortex for both intermittent (iTBS) and continuous (cTBS) TBS using machine learning. Validation of the created model showed an area under the curve (AUC) of 0.85 and 0.69 and positive predictive values of 77.7 and 70.0% for iTBS and cTBS, respectively; the negative predictive value was 75.5% for both patterns. Additionally, the accuracy was 0.76 and 0.72, precision was 0.82 and 0.67, recall was 0.82 and 0.67, and F1 scores were 0.82 and 0.67 for iTBS and cTBS, respectively. The most important predictor of iTBS was the motor evoked potential amplitude, whereas it was the intracortical facilitation for cTBS. Our results provide additional insights into the prediction of the effects of TBS variability according to baseline neurophysiological factors.

A primer on the use of machine learning to distil knowledge from data in biological psychiatry

Applications of machine learning in the biomedical sciences are growing rapidly. This growth has been spurred by diverse cross-institutional and interdisciplinary collaborations, public availability of large datasets, an increase in the accessibility of analytic routines, and the availability of powerful computing resources. With this increased access and exposure to machine learning comes a responsibility for education and a deeper understanding of its bases and bounds, borne equally by data scientists seeking to ply their analytic wares in medical research and by biomedical scientists seeking to harness such methods to glean knowledge from data. This article provides an accessible and critical review of machine learning for a biomedically informed audience, as well as its applications in psychiatry. The review covers definitions and expositions of commonly used machine learning methods, and historical trends of their use in psychiatry. We also provide a set of standards, namely Guidelines for REporting Machine Learning Investigations in Neuropsychiatry (GREMLIN), for designing and reporting studies that use machine learning as a primary data-analysis approach. Lastly, we propose the establishment of the Machine Learning in Psychiatry (MLPsych) Consortium, enumerate its objectives, and identify areas of opportunity for future applications of machine learning in biological psychiatry. This review serves as a cautiously optimistic primer on machine learning for those on the precipice as they prepare to dive into the field, either as methodological practitioners or well-informed consumers.

Electome network factors: Capturing emotional brain networks related to health and disease

Therapeutic development for mental disorders has been slow despite the high worldwide prevalence of illness. Unfortunately, cellular and circuit insights into disease etiology have largely failed to generalize across individuals that carry the same diagnosis, reflecting an unmet need to identify convergent mechanisms that would facilitate optimal treatment. Here, we discuss how mesoscale networks can encode affect and other cognitive processes. These networks can be discovered through electrical functional connectome (electome) analysis, a method built upon explainable machine learning models for analyzing and interpreting mesoscale brain-wide signals in a behavioral context. We also outline best practices for identifying these generalizable, interpretable, and biologically relevant networks. Looking forward, translational electome analysis can span species and various moods, cognitive processes, or other brain states, supporting translational medicine. Thus, we argue that electome analysis provides potential translational biomarkers for developing next-generation therapeutics that exhibit high efficacy across heterogeneous disorders.

Human behavior in free search online shopping scenarios can be predicted from EEG activation using Hjorth parameters

The present work investigates whether and how decisions in real-world online shopping scenarios can be predicted based on brain activation. Potential customers were asked to search through product pages on e-commerce platforms and decide, which products to buy, while their EEG signal was recorded. Machine learning algorithms were then trained to distinguish between EEG activation when viewing products that are later bought or put into the shopping card as opposed to products that are later discarded. We find that Hjorth parameters extracted from the raw EEG can be used to predict purchase choices to a high level of accuracy. Above-chance predictions based on Hjorth parameters are achieved via different standard machine learning methods with random forest models showing the best performance of above 80% prediction accuracy in both 2-class (bought or put into card vs. not bought) and 3-class (bought vs. put into card vs. not bought) classification. While conventional EEG signal analysis commonly employs frequency domain features such as alpha or theta power and phase, Hjorth parameters use time domain signals, which can be calculated rapidly with little computational cost. Given the presented evidence that Hjorth parameters are suitable for the prediction of complex behaviors, their potential and remaining challenges for implementation in real-time applications are discussed.

Dissociable default-mode subnetworks subserve childhood attention and cognitive flexibility: Evidence from deep learning and stereotactic electroencephalography

Cognitive flexibility encompasses the ability to efficiently shift focus and forms a critical component of goal-directed attention. The neural substrates of this process are incompletely understood in part due to difficulties in sampling the involved circuitry. We leverage stereotactic intracranial recordings to directly resolve local-field potentials from otherwise inaccessible structures to study moment-to-moment attentional activity in children with epilepsy performing a flexible attentional task. On an individual subject level, we employed deep learning to decode neural features predictive of task performance indexed by single-trial reaction time. These models were subsequently aggregated across participants to identify predictive brain regions based on AAL atlas and FIND functional network parcellations. Through this approach, we show that fluctuations in beta (12-30 Hz) and gamma (30-80 Hz) power reflective of increased top-down attentional control and local neuronal processing within relevant large-scale networks can accurately predict single-trial task performance. We next performed connectomic profiling of these highly predictive nodes to examine task-related engagement of distributed functional networks, revealing exclusive recruitment of the dorsal default mode network during shifts in attention. The identification of distinct substreams within the default mode system supports a key role for this network in cognitive flexibility and attention in children. Furthermore, convergence of our results onto consistent functional networks despite significant inter-subject variability in electrode implantations supports a broader role for deep learning applied to intracranial electrodes in the study of human attention.

Prediction of depression treatment outcome from multimodal data: a CAN-BIND-1 report

BACKGROUND

Prediction of treatment outcomes is a key step in improving the treatment of major depressive disorder (MDD). The Canadian Biomarker Integration Network in Depression (CAN-BIND) aims to predict antidepressant treatment outcomes through analyses of clinical assessment, neuroimaging, and blood biomarkers.

METHODS

In the CAN-BIND-1 dataset of 192 adults with MDD and outcomes of treatment with escitalopram, we applied machine learning models in a nested cross-validation framework. Across 210 analyses, we examined combinations of predictive variables from three modalities, measured at baseline and after 2 weeks of treatment, and five machine learning methods with and without feature selection. To optimize the predictors-to-observations ratio, we followed a tiered approach with 134 and 1152 variables in tier 1 and tier 2 respectively.

RESULTS

A combination of baseline tier 1 clinical, neuroimaging, and molecular variables predicted response with a mean balanced accuracy of 0.57 (best model mean 0.62) compared to 0.54 (best model mean 0.61) in single modality models. Adding week 2 predictors improved the prediction of response to a mean balanced accuracy of 0.59 (best model mean 0.66). Adding tier 2 features did not improve prediction.

CONCLUSIONS

A combination of clinical, neuroimaging, and molecular data improves the prediction of treatment outcomes over single modality measurement. The addition of measurements from the early stages of treatment adds precision. Present results are limited by lack of external validation. To achieve clinically meaningful prediction, the multimodal measurement should be scaled up to larger samples and the robustness of prediction tested in an external validation dataset.

Neuroscience Needs Network Science

The brain is a complex system comprising a myriad of interacting neurons, posing significant challenges in understanding its structure, function, and dynamics. Network science has emerged as a powerful tool for studying such interconnected systems, offering a framework for integrating multiscale data and complexity. To date, network methods have significantly advanced functional imaging studies of the human brain and have facilitated the development of control theory-based applications for directing brain activity. Here, we discuss emerging frontiers for network neuroscience in the brain atlas era, addressing the challenges and opportunities in integrating multiple data streams for understanding the neural transitions from development to healthy function to disease. We underscore the importance of fostering interdisciplinary opportunities through workshops, conferences, and funding initiatives, such as supporting students and postdoctoral fellows with interests in both disciplines. By bringing together the network science and neuroscience communities, we can develop novel network-based methods tailored to neural circuits, paving the way toward a deeper understanding of the brain and its functions, as well as offering new challenges for network science.

Explainable Artificial Intelligence (XAI): Concepts and Challenges in Healthcare

Artificial Intelligence (AI) describes computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and language translation. Examples of AI techniques are machine learning, neural networks, and deep learning. AI can be applied in many different areas, such as econometrics, biometry, e-commerce, and the automotive industry. In recent years, AI has found its way into healthcare as well, helping doctors make better decisions (“clinical decision support”), localizing tumors in magnetic resonance images, reading and analyzing reports written by radiologists and pathologists, and much more. However, AI has one big risk: it can be perceived as a “black box”, limiting trust in its reliability, which is a very big issue in an area in which a decision can mean life or death. As a result, the term Explainable Artificial Intelligence (XAI) has been gaining momentum. XAI tries to ensure that AI algorithms (and the resulting decisions) can be understood by humans. In this narrative review, we will have a look at some central concepts in XAI, describe several challenges around XAI in healthcare, and discuss whether it can really help healthcare to advance, for example, by increasing understanding and trust. Finally, alternatives to increase trust in AI are discussed, as well as future research possibilities in the area of XAI.

Decoding Mental Effort in a Quasi-Realistic Scenario: A Feasibility Study on Multimodal Data Fusion and Classification

Humans' performance varies due to the mental resources that are available to successfully pursue a task. To monitor users' current cognitive resources in naturalistic scenarios, it is essential to not only measure demands induced by the task itself but also consider situational and environmental influences. We conducted a multimodal study with 18 participants (nine female, M = 25.9 with SD = 3.8 years). In this study, we recorded respiratory, ocular, cardiac, and brain activity using functional near-infrared spectroscopy (fNIRS) while participants performed an adapted version of the warship commander task with concurrent emotional speech distraction. We tested the feasibility of decoding the experienced mental effort with a multimodal machine learning architecture. The architecture comprised feature engineering, model optimisation, and model selection to combine multimodal measurements in a cross-subject classification. Our approach reduces possible overfitting and reliably distinguishes two different levels of mental effort. These findings contribute to the prediction of different states of mental effort and pave the way toward generalised state monitoring across individuals in realistic applications.

The evolution of Big Data in neuroscience and neurology

Neurological diseases are on the rise worldwide, leading to increased healthcare costs and diminished quality of life in patients. In recent years, Big Data has started to transform the fields of Neuroscience and Neurology. Scientists and clinicians are collaborating in global alliances, combining diverse datasets on a massive scale, and solving complex computational problems that demand the utilization of increasingly powerful computational resources. This Big Data revolution is opening new avenues for developing innovative treatments for neurological diseases. Our paper surveys Big Data's impact on neurological patient care, as exemplified through work done in a comprehensive selection of areas, including Connectomics, Alzheimer's Disease, Stroke, Depression, Parkinson's Disease, Pain, and Addiction (e.g., Opioid Use Disorder). We present an overview of research and the methodologies utilizing Big Data in each area, as well as their current limitations and technical challenges. Despite the potential benefits, the full potential of Big Data in these fields currently remains unrealized. We close with recommendations for future research aimed at optimizing the use of Big Data in Neuroscience and Neurology for improved patient outcomes.

Supplementary Information

The online version contains supplementary material available at 10.1186/s40537-023-00751-2.

Differences in brain connectivity of meditators during assessing neurocognition via gamified experimental logic task: A machine learning approach

Meditation practices increase attention, memory, and self-awareness. The neuroscientific study of meditation has helped gain useful insights into the functional changes in the brain. In this study, we have assessed the performance of meditators with different years of practice while performing an engaging task rather than studying the meditation practice itself. This task helps assess many neural processes simultaneously and represents task performance in presence of multiple audio-visual distractors as in a real-life scenario. The long-term practice of meditation could bring neuroplastic changes in the way cognitive processing is carried out. It could be conscious and effortful in short-term practitioners and relatively unconscious and effortless in long-term practitioners. Our goal is to understand if it is possible to differentiate between long-term and short-term meditators solely based on their cognitive processing. A group of proficient Rajayoga meditators from the Brahma Kumaris were recruited based on their meditation experience-Long-Term Practitioners (n = 12, mean 13,596 h) and Short-Term Practitioners (n = 10, mean 1095 h). A task-based functional Magnetic Resonance Imaging was acquired while the subjects performed the task. Functional Connectivity Analysis was performed to derive the correlation measures to be used as features for classification. Five supervised Machine Learning algorithms Logistic Regression, Support Vector Machine, Decision Tree, Random Forest, and Gradient Boosted Tree were used for classification. Among all the classifiers Gradient Boosted Tree performed the best with an accuracy of 77% when all the four Functional Connectivity Metrics were used. Connectivity in visual areas, cerebellum, left rostral prefrontal cortex, and middle frontal gyrus was found to be higher in long-term meditators. Such a classification demonstrates that long-term meditation practice brings about neuroplastic changes that influence cognitive processing.

Neuroscience needs Network Science

The brain is a complex system comprising a myriad of interacting elements, posing significant challenges in understanding its structure, function, and dynamics. Network science has emerged as a powerful tool for studying such intricate systems, offering a framework for integrating multiscale data and complexity. Here, we discuss the application of network science in the study of the brain, addressing topics such as network models and metrics, the connectome, and the role of dynamics in neural networks. We explore the challenges and opportunities in integrating multiple data streams for understanding the neural transitions from development to healthy function to disease, and discuss the potential for collaboration between network science and neuroscience communities. We underscore the importance of fostering interdisciplinary opportunities through funding initiatives, workshops, and conferences, as well as supporting students and postdoctoral fellows with interests in both disciplines. By uniting the network science and neuroscience communities, we can develop novel network-based methods tailored to neural circuits, paving the way towards a deeper understanding of the brain and its functions.

External Validation of an Artificial Intelligence Device for Intracranial Hemorrhage Detection

BACKGROUND

Artificial intelligence applications have gained traction in the field of cerebrovascular disease by assisting in the triage, classification, and prognostication of both ischemic and hemorrhagic stroke. The Caire ICH system aims to be the first device to move into the realm of assisted diagnosis for intracranial hemorrhage (ICH) and its subtypes.

METHODS

A single-center retrospective dataset of 402 head noncontrast CT scans (NCCT) with an intracranial hemorrhage were retrospectively collected from January 2012 to July 2020; an additional 108 NCCT scans with no intracranial hemorrhage findings were also included. The presence of an ICH and its subtype were determined from the International Classification of Diseases-10 code associated with the scan and validated by an expert panel. We used the Caire ICH vR1 to analyze these scans, and we evaluated its performance in terms of accuracy, sensitivity, and specificity.

RESULTS

We found the Caire ICH system to have an accuracy of 98.05% (95% confidence interval [CI]: 96.44%-99.06%), a sensitivity of 97.52% (95% CI: 95.50%-98.81%), and a specificity of 100% (95% CI: 96.67%-100.00%) in the detection of ICH. Experts reviewed the 10 incorrectly classified scans.

CONCLUSIONS

The Caire ICH vR1 algorithm was highly accurate, sensitive, and specific in detecting the presence or absence of an ICH and its subtypes in NCCTs. This work suggests that the Caire ICH device has potential to minimize clinical errors in ICH diagnosis that could improve patient outcomes and current workflows as both a point-of-care tool for diagnostics and as a safety net for radiologists.

An overview of machine learning methods in enabling IoMT-based epileptic seizure detection

The healthcare industry is rapidly automating, in large part because of the Internet of Things (IoT). The sector of the IoT devoted to medical research is sometimes called the Internet of Medical Things (IoMT). Data collecting and processing are the fundamental components of all IoMT applications. Machine learning (ML) algorithms must be included into IoMT immediately due to the vast quantity of data involved in healthcare and the value that precise forecasts have. In today's world, together, IoMT, cloud services, and ML techniques have become effective tools for solving many problems in the healthcare sector, such as epileptic seizure monitoring and detection. One of the biggest hazards to people's lives is epilepsy, a lethal neurological condition that has become a global issue. To prevent the deaths of thousands of epileptic patients each year, there is a critical necessity for an effective method for detecting epileptic seizures at their earliest stage. Numerous medical procedures, including epileptic monitoring, diagnosis, and other procedures, may be carried out remotely with the use of IoMT, which will reduce healthcare expenses and improve services. This article seeks to act as both a collection and a review of the different cutting-edge ML applications for epilepsy detection that are presently being combined with IoMT.

From local counterfactuals to global feature importance: efficient, robust, and model-agnostic explanations for brain connectivity networks

BACKGROUND

Explainable artificial intelligence (XAI) is a technology that can enhance trust in mental state classifications by providing explanations for the reasoning behind artificial intelligence (AI) models outputs, especially for high-dimensional and highly-correlated brain signals. Feature importance and counterfactual explanations are two common approaches to generate these explanations, but both have drawbacks. While feature importance methods, such as shapley additive explanations (SHAP), can be computationally expensive and sensitive to feature correlation, counterfactual explanations only explain a single outcome instead of the entire model.

METHODS

To overcome these limitations, we propose a new procedure for computing global feature importance that involves aggregating local counterfactual explanations. This approach is specifically tailored to fMRI signals and is based on the hypothesis that instances close to the decision boundary and their counterfactuals mainly differ in the features identified as most important for the downstream classification task. We refer to this proposed feature importance measure as Boundary Crossing Solo Ratio (BoCSoR), since it quantifies the frequency with which a change in each feature in isolation leads to a change in classification outcome, i.e., the crossing of the model's decision boundary.

RESULTS AND CONCLUSIONS

Experimental results on synthetic data and real publicly available fMRI data from the Human Connect project show that the proposed BoCSoR measure is more robust to feature correlation and less computationally expensive than state-of-the-art methods. Additionally, it is equally effective in providing an explanation for the behavior of any AI model for brain signals. These properties are crucial for medical decision support systems, where many different features are often extracted from the same physiological measures and a gold standard is absent. Consequently, computing feature importance may become computationally expensive, and there may be a high probability of mutual correlation among features, leading to unreliable results from state-of-the-art XAI methods.

Inferring Pyramidal Neuron Morphology using EAP Data

We report a computational algorithm that uses an inverse modeling scheme to infer neuron position and morphology of cortical pyramidal neurons using spatio-temporal extracellular action potential recordings.. We first develop a generic pyramidal neuron model with stylized morphology and active channels that could mimic the realistic electrophysiological dynamics of pyramidal cells from different cortical layers. The generic stylized single neuron model has adjustable parameters for soma location, and morphology and orientation of the dendrites. The ranges for the parameters were selected to include morphology of the pyramidal neuron types in the rodent primary motor cortex. We then developed a machine learning approach that uses the local field potential simulated from the stylized model for training a convolutional neural network that predicts the parameters of the stylized neuron model. Preliminary results suggest that the proposed methodology can reliably infer the key position and morphology parameters using the simulated spatio-temporal profile of EAP waveforms. We also provide partial support to validate the inference algorithm using in vivo data. Finally, we highlight the issues involved and ongoing work to develop a pipeline to automate the scheme.

Explainable artificial intelligence for mental health through transparency and interpretability for understandability

The literature on artificial intelligence (AI) or machine learning (ML) in mental health and psychiatry lacks consensus on what "explainability" means. In the more general XAI (eXplainable AI) literature, there has been some convergence on explainability meaning model-agnostic techniques that augment a complex model (with internal mechanics intractable for human understanding) with a simpler model argued to deliver results that humans can comprehend. Given the differing usage and intended meaning of the term "explainability" in AI and ML, we propose instead to approximate model/algorithm explainability by understandability defined as a function of transparency and interpretability. These concepts are easier to articulate, to "ground" in our understanding of how algorithms and models operate and are used more consistently in the literature. We describe the TIFU (Transparency and Interpretability For Understandability) framework and examine how this applies to the landscape of AI/ML in mental health research. We argue that the need for understandablity is heightened in psychiatry because data describing the syndromes, outcomes, disorders and signs/symptoms possess probabilistic relationships to each other-as do the tentative aetiologies and multifactorial social- and psychological-determinants of disorders. If we develop and deploy AI/ML models, ensuring human understandability of the inputs, processes and outputs of these models is essential to develop trustworthy systems fit for deployment.

External validation of binary machine learning models for pain intensity perception classification from EEG in healthy individuals

Discrimination of pain intensity using machine learning (ML) and electroencephalography (EEG) has significant potential for clinical applications, especially in scenarios where self-report is unsuitable. However, existing research is limited due to a lack of external validation (assessing performance using novel data). We aimed for the first external validation study for pain intensity classification with EEG. Pneumatic pressure stimuli were delivered to the fingernail bed at high and low pain intensities during two independent EEG experiments with healthy participants. Study one (n = 25) was utilised for training and cross-validation. Study two (n = 15) was used for external validation one (identical stimulation parameters to study one) and external validation two (new stimulation parameters). Time-frequency features of peri-stimulus EEG were computed on a single-trial basis for all electrodes. ML training and analysis were performed on a subset of features, identified through feature selection, which were distributed across scalp electrodes and included frontal, central, and parietal regions. Results demonstrated that ML models outperformed chance. The Random Forest (RF) achieved the greatest accuracies of 73.18, 68.32 and 60.42% for cross-validation, external validation one and two, respectively. Importantly, this research is the first to externally validate ML and EEG for the classification of intensity during experimental pain, demonstrating promising performance which generalises to novel samples and paradigms. These findings offer the most rigorous estimates of ML's clinical potential for pain classification.

Supervising the Decoder of Variational Autoencoders to Improve Scientific Utility

Probabilistic generative models are attractive for scientific modeling because their inferred parameters can be used to generate hypotheses and design experiments. This requires that the learned model provides an accurate representation of the input data and yields a latent space that effectively predicts outcomes relevant to the scientific question. Supervised Variational Autoencoders (SVAEs) have previously been used for this purpose, as a carefully designed decoder can be used as an interpretable generative model of the data, while the supervised objective ensures a predictive latent representation. Unfortunately, the supervised objective forces the encoder to learn a biased approximation to the generative posterior distribution, which renders the generative parameters unreliable when used in scientific models. This issue has remained undetected as reconstruction losses commonly used to evaluate model performance do not detect bias in the encoder. We address this previously-unreported issue by developing a second-order supervision framework (SOS-VAE) that updates the decoder parameters, rather than the encoder, to induce a predictive latent representation. This ensures that the encoder maintains a reliable posterior approximation and the decoder parameters can be effectively interpreted. We extend this technique to allow the user to trade-off the bias in the generative parameters for improved predictive performance, acting as an intermediate option between SVAEs and our new SOS-VAE. We also use this methodology to address missing data issues that often arise when combining recordings from multiple scientific experiments. We demonstrate the effectiveness of these developments using synthetic data and electrophysiological recordings with an emphasis on how our learned representations can be used to design scientific experiments.

Integrating Statistical and Machine Learning Approaches for Neural Classification

Neurons can code for multiple variables simultaneously and neuroscientists are often interested in classifying neurons based on their receptive field properties. Statistical models provide powerful tools for determining the factors influencing neural spiking activity and classifying individual neurons. However, as neural recording technologies have advanced to produce simultaneous spiking data from massive populations, classical statistical methods often lack the computational efficiency required to handle such data. Machine learning (ML) approaches are known for enabling efficient large scale data analyses; however, they typically require massive training sets with balanced data, along with accurate labels to fit well. Additionally, model assessment and interpretation are often more challenging for ML than for classical statistical methods. To address these challenges, we develop an integrated framework, combining statistical modeling and machine learning approaches to identify the coding properties of neurons from large populations. In order to demonstrate this framework, we apply these methods to data from a population of neurons recorded from rat hippocampus to characterize the distribution of spatial receptive fields in this region.

Italian, European, and international neuroinformatics efforts: An overview

Neuroinformatics is a research field that focusses on software tools capable of identifying, analysing, modelling, organising and sharing multiscale neuroscience data. Neuroinformatics has exploded in the last two decades with the emergence of the Big Data phenomenon, characterised by the so-called 3Vs (volume, velocity and variety), which provided neuroscientists with an improved ability to acquire and process data faster and more cheaply thanks to technical improvements in clinical, genomic and radiological technologies. This situation has led to a 'data deluge', as neuroscientists can routinely collect more study data in a few days than they could in a year just a decade ago. To address this phenomenon, several neuroimaging-focussed neuroinformatics platforms have emerged, funded by national or transnational agencies, with the following goals: (i) development of tools for archiving and organising analytical data (XNAT, REDCap and LabKey); (ii) development of data-driven models evolving from reductionist approaches to multidimensional models (RIN, IVN, HBD, EuroPOND, E-DADS and GAAIN BRAIN); and (iii) development of e-infrastructures to provide sufficient computational power and storage resources (neuGRID, HBP-EBRAINS, LONI and CONP). Although the scenario is still fragmented, there are technological and economical attempts at both national and international levels to introduce high standards for open and Findable, Accessible, Interoperable and Reusable (FAIR) neuroscience worldwide.

ABOT: an open-source online benchmarking tool for machine learning-based artefact detection and removal methods from neuronal signals

Brain signals are recorded using different techniques to aid an accurate understanding of brain function and to treat its disorders. Untargeted internal and external sources contaminate the acquired signals during the recording process. Often termed as artefacts, these contaminations cause serious hindrances in decoding the recorded signals; hence, they must be removed to facilitate unbiased decision-making for a given investigation. Due to the complex and elusive manifestation of artefacts in neuronal signals, computational techniques serve as powerful tools for their detection and removal. Machine learning (ML) based methods have been successfully applied in this task. Due to ML's popularity, many articles are published every year, making it challenging to find, compare and select the most appropriate method for a given experiment. To this end, this paper presents ABOT (Artefact removal Benchmarking Online Tool) as an online benchmarking tool which allows users to compare existing ML-driven artefact detection and removal methods from the literature. The characteristics and related information about the existing methods have been compiled as a knowledgebase (KB) and presented through a user-friendly interface with interactive plots and tables for users to search it using several criteria. Key characteristics extracted from over 120 articles from the literature have been used in the KB to help compare the specific ML models. To comply with the FAIR (Findable, Accessible, Interoperable and Reusable) principle, the source code and documentation of the toolbox have been made available via an open-access repository.

Spontaneous Activity Predicts Survival of Developing Cortical Neurons

Spontaneous activity plays a crucial role in brain development by coordinating the integration of immature neurons into emerging cortical networks. High levels and complex patterns of spontaneous activity are generally associated with low rates of apoptosis in the cortex. However, whether spontaneous activity patterns directly encode for survival of individual cortical neurons during development remains an open question. Here, we longitudinally investigated spontaneous activity and apoptosis in developing cortical cultures, combining extracellular electrophysiology with calcium imaging. These experiments demonstrated that the early occurrence of calcium transients was strongly linked to neuronal survival. Silent neurons exhibited a higher probability of cell death, whereas high frequency spiking and burst behavior were almost exclusively detected in surviving neurons. In local neuronal clusters, activity of neighboring neurons exerted a pro-survival effect, whereas on the functional level, networks with a high modular topology were associated with lower cell death rates. Using machine learning algorithms, cell fate of individual neurons was predictable through the integration of spontaneous activity features. Our results indicate that high frequency spiking activity constrains apoptosis in single neurons through sustained calcium rises and thereby consolidates networks in which a high modular topology is reached during early development.

A robust framework to investigate the reliability and stability of explainable artificial intelligence markers of Mild Cognitive Impairment and Alzheimer's Disease

In clinical practice, several standardized neuropsychological tests have been designed to assess and monitor the neurocognitive status of patients with neurodegenerative diseases such as Alzheimer's disease. Important research efforts have been devoted so far to the development of multivariate machine learning models that combine the different test indexes to predict the diagnosis and prognosis of cognitive decline with remarkable results. However, less attention has been devoted to the explainability of these models. In this work, we present a robust framework to (i) perform a threefold classification between healthy control subjects, individuals with cognitive impairment, and subjects with dementia using different cognitive indexes and (ii) analyze the variability of the explainability SHAP values associated with the decisions taken by the predictive models. We demonstrate that the SHAP values can accurately characterize how each index affects a patient's cognitive status. Furthermore, we show that a longitudinal analysis of SHAP values can provide effective information on Alzheimer's disease progression.

From End to End: Gaining, Sorting, and Employing High-Density Neural Single Unit Recordings

The meaning behind neural single unit activity has constantly been a challenge, so it will persist in the foreseeable future. As one of the most sourced strategies, detecting neural activity in high-resolution neural sensor recordings and then attributing them to their corresponding source neurons correctly, namely the process of spike sorting, has been prevailing so far. Support from ever-improving recording techniques and sophisticated algorithms for extracting worthwhile information and abundance in clustering procedures turned spike sorting into an indispensable tool in electrophysiological analysis. This review attempts to illustrate that in all stages of spike sorting algorithms, the past 5 years innovations' brought about concepts, results, and questions worth sharing with even the non-expert user community. By thoroughly inspecting latest innovations in the field of neural sensors, recording procedures, and various spike sorting strategies, a skeletonization of relevant knowledge lays here, with an initiative to get one step closer to the original objective: deciphering and building in the sense of neural transcript.

Brain-wide electrical dynamics encode individual appetitive social behavior

The architecture whereby activity across many brain regions integrates to encode individual appetitive social behavior remains unknown. Here we measure electrical activity from eight brain regions as mice engage in a social preference assay. We then use machine learning to discover a network that encodes the extent to which individual mice engage another mouse. This network is organized by theta oscillations leading from prelimbic cortex and amygdala that converge on the ventral tegmental area. Network activity is synchronized with cellular firing, and frequency-specific activation of a circuit within this network increases social behavior. Finally, the network generalizes, on a mouse-by-mouse basis, to encode individual differences in social behavior in healthy animals but fails to encode individual behavior in a 'high confidence' genetic model of autism. Thus, our findings reveal the architecture whereby the brain integrates distributed activity across timescales to encode an appetitive brain state underlying individual differences in social behavior.

Oscillatory brain activity links experience to expectancy during associative learning

Associating a novel situation with a specific outcome involves a cascade of cognitive processes, including selecting relevant stimuli, forming predictions regarding expected outcomes, and updating memorized predictions based on experience. The present manuscript uses computational modeling and machine learning to test the hypothesis that alpha-band (8-12 Hz) oscillations are involved in the updating of expectations based on experience. Participants learned that a visual cue predicted an aversive loud noise with a probability of 50%. The Rescorla-Wagner model of associative learning explained trial-wise changes in self-reported noise expectancy as well as alpha power changes. Experience in the past trial and self-reported expectancy for the subsequent trial were accurately decoded based on the topographical distribution of alpha power at specific latencies. Decodable information during initial association formation and contingency report recurred when viewing the conditioned cue. Findings support the idea that alpha oscillations have multiple, temporally specific, roles in the formation of associations between cues and outcomes.

A Machine Learning Approach for Detecting Cognitive Interference Based on Eye-Tracking Data

The Stroop test evaluates the ability to inhibit cognitive interference. This interference occurs when the processing of one stimulus characteristic affects the simultaneous processing of another attribute of the same stimulus. Eye movements are an indicator of the individual attention load required for inhibiting cognitive interference. We used an eye tracker to collect eye movements data from more than 60 subjects each performing four different but similar tasks (some with cognitive interference and some without). After the extraction of features related to fixations, saccades and gaze trajectory, we trained different Machine Learning models to recognize tasks performed in the different conditions (i.e., with interference, without interference). The models achieved good classification performances when distinguishing between similar tasks performed with or without cognitive interference. This suggests the presence of characterizing patterns common among subjects, which can be captured by machine learning algorithms despite the individual variability of visual behavior.

Time-Frequency Representations of Brain Oscillations: Which One Is Better?

Brain oscillations are thought to subserve important functions by organizing the dynamical landscape of neural circuits. The expression of such oscillations in neural signals is usually evaluated using time-frequency representations (TFR), which resolve oscillatory processes in both time and frequency. While a vast number of methods exist to compute TFRs, there is often no objective criterion to decide which one is better. In feature-rich data, such as that recorded from the brain, sources of noise and unrelated processes abound and contaminate results. The impact of these distractor sources is especially problematic, such that TFRs that are more robust to contaminants are expected to provide more useful representations. In addition, the minutiae of the techniques themselves impart better or worse time and frequency resolutions, which also influence the usefulness of the TFRs. Here, we introduce a methodology to evaluate the "quality" of TFRs of neural signals by quantifying how much information they retain about the experimental condition during visual stimulation and recognition tasks, in mice and humans, respectively. We used machine learning to discriminate between various experimental conditions based on TFRs computed with different methods. We found that various methods provide more or less informative TFRs depending on the characteristics of the data. In general, however, more advanced techniques, such as the superlet transform, seem to provide better results for complex time-frequency landscapes, such as those extracted from electroencephalography signals. Finally, we introduce a method based on feature perturbation that is able to quantify how much time-frequency components contribute to the correct discrimination among experimental conditions. The methodology introduced in the present study may be extended to other analyses of neural data, enabling the discovery of data features that are modulated by the experimental manipulation.

Toward the explainability, transparency, and universality of machine learning for behavioral classification in neuroscience

The use of rigorous ethological observation via machine learning techniques to understand brain function (computational neuroethology) is a rapidly growing approach that is poised to significantly change how behavioral neuroscience is commonly performed. With the development of open-source platforms for automated tracking and behavioral recognition, these approaches are now accessible to a wide array of neuroscientists despite variations in budget and computational experience. Importantly, this adoption has moved the field toward a common understanding of behavior and brain function through the removal of manual bias and the identification of previously unknown behavioral repertoires. Although less apparent, another consequence of this movement is the introduction of analytical tools that increase the explainabilty, transparency, and universality of the machine-based behavioral classifications both within and between research groups. Here, we focus on three main applications of such machine model explainabilty tools and metrics in the drive toward behavioral (i) standardization, (ii) specialization, and (iii) explainability. We provide a perspective on the use of explainability tools in computational neuroethology, and detail why this is a necessary next step in the expansion of the field. Specifically, as a possible solution in behavioral neuroscience, we propose the use of Shapley values via Shapley Additive Explanations (SHAP) as a diagnostic resource toward explainability of human annotation, as well as supervised and unsupervised behavioral machine learning analysis.

The Nencki-Symfonia electroencephalography/event-related potential dataset: Multiple cognitive tasks and resting-state data collected in a sample of healthy adults

BACKGROUND

One of the goals of neuropsychology is to understand the brain mechanisms underlying aspects of attention and cognitive control. Several tasks have been developed as a part of this body of research, however their results are not always consistent. A reliable comparison of the data and a synthesis of study conclusions has been precluded by multiple methodological differences. Here, we describe a publicly available, high-density electroencephalography (EEG) dataset obtained from 42 healthy young adults while they performed 3 cognitive tasks: (i) an extended multi-source interference task; (ii) a 3-stimuli oddball task; (iii) a control, simple reaction task; and (iv) a resting-state protocol. Demographic and psychometric information are included within the dataset.

DATASET VALIDATION

First, data validation confirmed acceptable quality of the obtained EEG signals. Typical event-related potential (ERP) waveforms were obtained, as expected for attention and cognitive control tasks (i.e., N200, P300, N450). Behavioral results showed the expected progression of reaction times and error rates, which confirmed the effectiveness of the applied paradigms.

CONCLUSIONS

This dataset is well suited for neuropsychological research regarding common and distinct mechanisms involved in different cognitive tasks. Using this dataset, researchers can compare a wide range of classical EEG/ERP features across tasks for any selected subset of electrodes. At the same time, 128-channel EEG recording allows for source localization and detailed connectivity studies. Neurophysiological measures can be correlated with additional psychometric data obtained from the same participants. This dataset can also be used to develop and verify novel analytical and classification approaches that can advance the field of deep/machine learning algorithms, recognition of single-trial ERP responses to different task conditions, and detection of EEG/ERP features for use in brain-computer interface applications.

Voice in Parkinson's Disease: A Machine Learning Study

Introduction

Parkinson's disease (PD) is characterized by specific voice disorders collectively termed hypokinetic dysarthria. We here investigated voice changes by using machine learning algorithms, in a large cohort of patients with PD in different stages of the disease, OFF and ON therapy.

Methods

We investigated 115 patients affected by PD (mean age: 68.2 ± 9.2 years) and 108 age-matched healthy subjects (mean age: 60.2 ± 11.0 years). The PD cohort included 57 early-stage patients (Hoehn &Yahr ≤ 2) who never took L-Dopa for their disease at the time of the study, and 58 mid-advanced-stage patients (Hoehn &Yahr >2) who were chronically-treated with L-Dopa. We clinically evaluated voices using specific subitems of the Unified Parkinson's Disease Rating Scale and the Voice Handicap Index. Voice samples recorded through a high-definition audio recorder underwent machine learning analysis based on the support vector machine classifier. We also calculated the receiver operating characteristic curves to examine the diagnostic accuracy of the analysis and assessed possible clinical-instrumental correlations.

Results

Voice is abnormal in early-stage PD and as the disease progresses, voice increasingly degradres as demonstrated by high accuracy in the discrimination between healthy subjects and PD patients in the early-stage and mid-advanced-stage. Also, L-dopa therapy improves but not restore voice in PD as shown by high accuracy in the comparison between patients OFF and ON therapy. Finally, for the first time we achieved significant clinical-instrumental correlations by using a new score (LR value) calculated by machine learning.

Conclusion

Voice is abnormal in early-stage PD, progressively degrades in mid-advanced-stage and can be improved but not restored by L-Dopa. Lastly, machine learning allows tracking disease severity and quantifying the symptomatic effect of L-Dopa on voice parameters with previously unreported high accuracy, thus representing a potential new biomarker of PD.

Alteration of Neural Network Activity With Aging Focusing on Temporal Complexity and Functional Connectivity Within Electroencephalography

With the aging process, brain functions, such as attention, memory, and cognitive functions, degrade over time. In a super-aging society, the alteration of neural activity owing to aging is considered crucial for interventions for the prevention of brain dysfunction. The complexity of temporal neural fluctuations with temporal scale dependency plays an important role in optimal brain information processing, such as perception and thinking. Complexity analysis is a useful approach for detecting cortical alteration in healthy individuals, as well as in pathological conditions, such as senile psychiatric disorders, resulting in changes in neural activity interactions among a wide range of brain regions. Multi-fractal (MF) and multi-scale entropy (MSE) analyses are known methods for capturing the complexity of temporal scale dependency of neural activity in the brain. MF and MSE analyses exhibit high accuracy in detecting changes in neural activity and are superior with regard to complexity detection when compared with other methods. In addition to complex temporal fluctuations, functional connectivity reflects the integration of information of brain processes in each region, described as mutual interactions of neural activity among brain regions. Thus, we hypothesized that the complementary relationship between functional connectivity and complexity could improve the ability to detect the alteration of spatiotemporal patterns observed on electroencephalography (EEG) with respect to aging. To prove this hypothesis, this study investigated the relationship between the complexity of neural activity and functional connectivity in aging based on EEG findings. Concretely, MF and MSE analyses were performed to evaluate the temporal complexity profiles, and phase lag index analyses assessing the unique profile of functional connectivity were performed based on the EEGs conducted for young and older participants. Subsequently, these profiles were combined through machine learning. We found that the complementary relationship between complexity and functional connectivity improves the classification accuracy among aging participants. Thus, the outcome of this study could be beneficial in formulating interventions for the prevention of age-related brain dysfunction.

Feeding the machine: Challenges to reproducible predictive modeling in resting-state connectomics

In this critical review, we examine the application of predictive models, for example, classifiers, trained using machine learning (ML) to assist in interpretation of functional neuroimaging data. Our primary goal is to summarize how ML is being applied and critically assess common practices. Our review covers 250 studies published using ML and resting-state functional MRI (fMRI) to infer various dimensions of the human functional connectome. Results for holdout ("lockbox") performance was, on average, ∼13% less accurate than performance measured through cross-validation alone, highlighting the importance of lockbox data, which was included in only 16% of the studies. There was also a concerning lack of transparency across the key steps in training and evaluating predictive models. The summary of this literature underscores the importance of the use of a lockbox and highlights several methodological pitfalls that can be addressed by the imaging community. We argue that, ideally, studies are motivated both by the reproducibility and generalizability of findings as well as the potential clinical significance of the insights. We offer recommendations for principled integration of machine learning into the clinical neurosciences with the goal of advancing imaging biomarkers of brain disorders, understanding causative determinants for health risks, and parsing heterogeneous patient outcomes.

Deep learning-based brain transcriptomic signatures associated with the neuropathological and clinical severity of Alzheimer's disease

Brain tissue gene expression from donors with and without Alzheimer's disease has been used to help inform the molecular changes associated with the development and potential treatment of this disorder. Here, we use a deep learning method to analyse RNA-seq data from 1114 brain donors from the Accelerating Medicines Project for Alzheimer's Disease consortium to characterize post-mortem brain transcriptome signatures associated with amyloid-β plaque, tau neurofibrillary tangles and clinical severity in multiple Alzheimer's disease dementia populations. Starting from the cross-sectional data in the Religious Orders Study and Memory and Aging Project cohort (n = 634), a deep learning framework was built to obtain a trajectory that mirrors Alzheimer's disease progression. A severity index was defined to quantitatively measure the progression based on the trajectory. Network analysis was then carried out to identify key gene (index gene) modules present in the model underlying the progression. Within this data set, severity indexes were found to be very closely correlated with all Alzheimer's disease neuropathology biomarkers (R ∼ 0.5, P < 1e-11) and global cognitive function (R = -0.68, P < 2.2e-16). We then applied the model to additional transcriptomic data sets from different brain regions (MAYO, n = 266; Mount Sinai Brain Bank, n = 214), and observed that the model remained significantly predictive (P < 1e-3) of neuropathology and clinical severity. The index genes that significantly contributed to the model were integrated with Alzheimer's disease co-expression regulatory networks, resolving four discrete gene modules that are implicated in vascular and metabolic dysfunction in different cell types, respectively. Our work demonstrates the generalizability of this signature to frontal and temporal cortex measurements and additional brain donors with Alzheimer's disease, other age-related neurological disorders and controls, and revealed that the transcriptomic network modules contribute to neuropathological and clinical disease severity. This study illustrates the promise of using deep learning methods to analyse heterogeneous omics data and discover potentially targetable molecular networks that can inform the development, treatment and prevention of neurodegenerative diseases like Alzheimer's disease.

Structural and functional brain networks of individual differences in trait anger and anger control: An unsupervised machine learning study

The ability to experience, use and eventually control anger is crucial to maintain well-being and build healthy relationships. Despite its relevance, the neural mechanisms behind individual differences in experiencing and controlling anger are poorly understood. To elucidate these points, we employed an unsupervised machine learning approach based on independent component analysis to test the hypothesis that specific functional and structural networks are associated with individual differences in trait anger and anger control. Structural and functional resting state images of 71 subjects as well as their scores from the State-Trait Anger Expression Inventory entered the analyses. At a structural level, the concentration of grey matter in a network including ventromedial temporal areas, posterior cingulate, fusiform gyrus and cerebellum was associated with trait anger. The higher the concentration, the higher the proneness to experience anger in daily life due to the greater tendency to orient attention towards aversive events and interpret them with higher hostility. At a functional level, the activity of the default mode network (DMN) was associated with anger control. The higher the DMN temporal frequency, the stronger the exerted control over anger, thus extending previous evidence on the role of the DMN in regulating cognitive and emotional functions in the domain of anger. Taken together, these results show, for the first time, two specialized brain networks for encoding individual differences in trait anger and anger control.

Machine Learning in Neuro-Oncology, Epilepsy, Alzheimer's Disease, and Schizophrenia

Applications of machine learning (ML) in translational medicine include therapeutic drug creation, diagnostic development, surgical planning, outcome prediction, and intraoperative assistance. Opportunities in the neurosciences are rich given advancement in our understanding of the brain, expanding indications for intervention, and diagnostic challenges often characterized by multiple clinical and environmental factors. We present a review of ML in neuro-oncology, epilepsy, Alzheimer's disease, and schizophrenia to highlight recent progression in these field, optimizing machine learning capabilities in their current forms. Supervised learning models appear to be the most commonly incorporated algorithm models for machine learning across the reviewed neuroscience disciplines with primary aim of diagnosis. Accuracy ranges are high from 63% to 99% across all algorithms investigated. Machine learning contributions to neurosurgery, neurology, psychiatry, and the clinical and basic science neurosciences may enhance current medical best practices while also broadening our understanding of dynamic neural networks and the brain.

Directed Spectral Measures Improve Latent Network Models Of Neural Populations

Systems neuroscience aims to understand how networks of neurons distributed throughout the brain mediate computational tasks. One popular approach to identify those networks is to first calculate measures of neural activity (e.g. power spectra) from multiple brain regions, and then apply a linear factor model to those measures. Critically, despite the established role of directed communication between brain regions in neural computation, measures of directed communication have been rarely utilized in network estimation because they are incompatible with the implicit assumptions of the linear factor model approach. Here, we develop a novel spectral measure of directed communication called the Directed Spectrum (DS). We prove that it is compatible with the implicit assumptions of linear factor models, and we provide a method to estimate the DS. We demonstrate that latent linear factor models of DS measures better capture underlying brain networks in both simulated and real neural recording data compared to available alternatives. Thus, linear factor models of the Directed Spectrum offer neuroscientists a simple and effective way to explicitly model directed communication in networks of neural populations.

Early prediction of developing spontaneous activity in cultured neuronal networks

Synchronization and bursting activity are intrinsic electrophysiological properties of in vivo and in vitro neural networks. During early development, cortical cultures exhibit a wide repertoire of synchronous bursting dynamics whose characterization may help to understand the parameters governing the transition from immature to mature networks. Here we used machine learning techniques to characterize and predict the developing spontaneous activity in mouse cortical neurons on microelectrode arrays (MEAs) during the first three weeks in vitro. Network activity at three stages of early development was defined by 18 electrophysiological features of spikes, bursts, synchrony, and connectivity. The variability of neuronal network activity during early development was investigated by applying k-means and self-organizing map (SOM) clustering analysis to features of bursts and synchrony. These electrophysiological features were predicted at the third week in vitro with high accuracy from those at earlier times using three machine learning models: Multivariate Adaptive Regression Splines, Support Vector Machines, and Random Forest. Our results indicate that initial patterns of electrical activity during the first week in vitro may already predetermine the final development of the neuronal network activity. The methodological approach used here may be applied to explore the biological mechanisms underlying the complex dynamics of spontaneous activity in developing neuronal cultures.