Quantifying performance of machine learning methods for neuroimaging data

Statistical agnostic regression: A machine learning method to validate regression models

INTRODUCTION

Regression analysis is a central topic in statistical modeling, aimed at estimating the relationships between a dependent variable, commonly referred to as the response variable, and one or more independent variables, i.e., explanatory variables. Linear regression is by far the most popular method for performing this task in various fields of research, such as data integration and predictive modeling when combining information from multiple sources.

OBJECTIVES

Classical methods for solving linear regression problems, such as Ordinary Least Squares (OLS), Ridge, or Lasso regressions, often form the foundation for more advanced machine learning (ML) techniques, which have been successfully applied, though without a formal definition of statistical significance. At most, permutation or analyses based on empirical measures (e.g., residuals or accuracy) have been conducted, leveraging the greater sensitivity of ML estimations for detection.

METHODS

In this paper, we introduce Statistical Agnostic Regression (SAR) for evaluating the statistical significance of ML-based linear regression models. This is achieved by analyzing concentration inequalities of the actual risk (expected loss) and considering the worst-case scenario. To this end, we define a threshold that ensures there is sufficient evidence, with a probability of at least 1-η, to conclude the existence of a linear relationship in the population between the explanatory (feature) and the response (label) variables.

CONCLUSIONS

Simulations demonstrate that the proposed agnostic (non-parametric) test can perform an analysis of variance comparable to the classical multivariate F-test for the slope parameter, without relying on the underlying assumptions of classical methods. A power analysis on a putative regression task revealed an overinflated false positive rate in standard ML methods, whereas the SAR test exhibited excellent control. Moreover, the residuals computed using this method represent a trade-off between those obtained from ML approaches and classical OLS.

Fingerprint patterns of human brain activity reveal a dynamic mix of emotional responses during virtual intergroup encounters

The Stereotype Content Model (SCM) states that different social groups elicit different emotions according to their perceived level of competence and warmth. Because of this relationship between stereotypes and emotional states and because emotions are highly predictive of intergroup behaviors, emotional evaluation is crucial for research on intergroup relations. However, emotional assessment heavily relies on self-reports, which are often compromised by social desirability and challenges in reporting immediate emotional appraisals. In this study, we used machine learning to identify emotional brain patterns using functional magnetic resonance imaging. Subsequently, those patterns were used to monitor emotional reactions during virtual intergroup encounters. Specifically, we showed Finnish majority group members 360-videos depicting members of their ethnic ingroup and immigrant outgroups approaching and entering participants' personal space. All the groups showed different levels of perceived competence and warmth. In alignment with the SCM, our results showed that the groups perceived as low in competence and warmth evoked contempt and discomfort. Moreover, the ambivalent low-competent/high-warm group elicited both happiness and discomfort. Additionally, upon the protagonists' approach into personal space, emotional reactions were modulated differently for each group. Taken together, our findings suggest that our method could be used to explore the temporal dynamics of emotional responses during intergroup encounters.

Prediction of Verbal Abilities From Brain Connectivity Data Across the Lifespan Using a Machine Learning Approach

Compared to nonverbal cognition such as executive or memory functions, language-related cognition generally appears to remain more stable until later in life. Nevertheless, different language-related processes, for example, verbal fluency versus vocabulary knowledge, appear to show different trajectories across the life span. One potential explanation for differences in verbal functions may be alterations in the functional and structural network architecture of different large-scale brain networks. For example, differences in verbal abilities have been linked to the communication within and between the frontoparietal (FPN) and default mode network (DMN). It, however, remains open whether brain connectivity within these networks may be informative for language performance at the individual level across the life span. Further information in this regard may be highly desirable as verbal abilities allow us to participate in daily activities, are associated with quality of life, and may be considered in preventive and interventional setups to foster cognitive health across the life span. So far, mixed prediction results based on resting-state functional connectivity (FC) and structural connectivity (SC) data have been reported for language abilities across different samples, age groups, and machine-learning (ML) approaches. Therefore, the current study set out to investigate the predictability of verbal fluency and vocabulary knowledge based on brain connectivity data in the DMN, FPN, and the whole brain using an ML approach in a lifespan sample (N = 717; age range: 18-85) from the 1000BRAINS study. Prediction performance was, thereby, systematically compared across (i) verbal [verbal fluency and vocabulary knowledge] and nonverbal abilities [processing speed and visual working memory], (ii) modalities [FC and SC data], (iii) feature sets [DMN, FPN, DMN-FPN, and whole brain], and (iv) samples [total, younger, and older aged group]. Results from the current study showed that verbal abilities could not be reliably predicted from FC and SC data across feature sets and samples. Thereby, no predictability differences emerged between verbal fluency and vocabulary knowledge across input modalities, feature sets, and samples. In contrast to verbal functions, nonverbal abilities could be moderately predicted from connectivity data, particularly SC, in the total and younger age group. Satisfactory prediction performance for nonverbal cognitive functions based on currently chosen connectivity data was, however, not encountered in the older age group. Current results, hence, emphasized that verbal functions may be more difficult to predict from brain connectivity data in domain-general cognitive networks and the whole brain compared to nonverbal abilities, particularly executive functions, across the life span. Thus, it appears warranted to more closely investigate differences in predictability between different cognitive functions and age groups.

Leveraging complementary multi-omics data integration methods for mechanistic insights in kidney diseases

Chronic kidney diseases (CKDs) are a global health concern, necessitating a comprehensive understanding of their complex pathophysiology. This study explores the use of 2 complementary multidimensional -omics data integration methods to elucidate mechanisms of CKD progression as a proof of concept. Baseline biosamples from 37 participants with CKD in the Clinical Phenotyping and Resource Biobank Core (C-PROBE) cohort with prospective longitudinal outcome data ascertained over 5 years were used to generate molecular profiles. Tissue transcriptomic, urine and plasma proteomic, and targeted urine metabolomic profiling were integrated using 2 orthogonal multi-omics data integration approaches, one unsupervised and the other supervised. Both integration methods identified 8 urinary proteins significantly associated with long-term outcomes, which were replicated in an adjusted survival model using 94 samples from an independent validation group in the same cohort. The 2 methods also identified 3 shared enriched pathways: the complement and coagulation cascades, cytokine-cytokine receptor interaction pathway, and the JAK/STAT signaling pathway. Use of different multiscalar data integration strategies on the same data enabled identification and prioritization of disease mechanisms associated with CKD progression. Approaches like this will be invaluable with the expansion of high-dimension data in kidney diseases.

Can artificial intelligence be the future solution to the enormous challenges and suffering caused by Schizophrenia?

This study evaluated the potential of artificial intelligence (AI) in the diagnosis, treatment, and prognostic assessment of schizophrenia (SZ) and explored collaborative directions for AI applications in future medical innovations. SZ is a severe mental disorder that causes significant suffering and imposes challenges on patients. With the rapid advancement of machine learning and deep learning technologies, AI has demonstrated notable advantages in the early diagnosis of high-risk populations. By integrating multidimensional biomarkers and linguistic behavior data of patients, AI can provide further objective and precise diagnostic criteria. Moreover, it aids in formulating personalized treatment plans, enhancing therapeutic outcomes, and offering new therapeutic strategies for patients with treatment-resistant SZ. Furthermore, AI excels in developing individualized prognostic plans, which enables the rapid identification of disease progression, accurate prediction of disease trajectory, and timely adjustment of treatment strategies, thereby improving prognosis and facilitating recovery. Despite the immense potential of AI in SZ management, its role as an auxiliary tool must be emphasized, with clinical judgment and compassionate care from healthcare professionals remaining crucial. Future research should focus on optimizing human-machine interactions to achieve efficient AI application in SZ management. The in-depth integration of AI technology into clinical practice will advance the field of SZ, ultimately improving the quality of life and treatment outcomes of patients.

Applications of MRI in Schizophrenia: Current Progress in Establishing Clinical Utility

Schizophrenia is a severe mental illness that significantly impacts the lives of affected individuals and with increasing mortality rates. Early detection and intervention are crucial for improving outcomes but the lack of validated biomarkers poses great challenges in such efforts. The use of magnetic resonance imaging (MRI) in schizophrenia enables the investigation of the disorder's etiological and neuropathological substrates in vivo. After decades of research, promising findings of MRI have been shown to aid in screening high-risk individuals and predicting illness onset, and predicting symptoms and treatment outcomes of schizophrenia. The integration of machine learning and deep learning techniques makes it possible to develop intelligent diagnostic and prognostic tools with extracted or selected imaging features. In this review, we aimed to provide an overview of current progress and prospects in establishing clinical utility of MRI in schizophrenia. We first provided an overview of MRI findings of brain abnormalities that might underpin the symptoms or treatment response process in schizophrenia patients. Then, we summarized the ongoing efforts in the computer-aided utility of MRI in schizophrenia and discussed the gap between MRI research findings and real-world applications. Finally, promising pathways to promote clinical translation were provided. EVIDENCE LEVEL: 2 TECHNICAL EFFICACY: Stage 3.

Application of Mechanistic Models and the Gaussian Process Model to Predict Bacterial Growth on Baby Spinach During Refrigerated Storage

Models that predict bacterial growth in food products can help industry with decision-making with regard to microbial food spoilage. Such models have recently been developed using machine learning (ML) rather than a mechanistic understanding of bacterial growth. Thus, our aim was to compare the performance of mechanistic (M) models and the Gaussian process (GP) model (i.e., an ML approach) for predicting bacterial growth on spinach from a US-based supply chain as well as a China-based supply chain; models were developed using previously published data, as well as new data collected in this study from the China-based supply chain. For the packaged spinach collected in this study from the China-based supply chain, the mean net growth of aerobic, mesophilic bacteria over 10 days of shelf life was 1.16 log10 (n = 11, local distribution) and 1.29 log10 (n = 8, eCommerce distribution); bacterial growth on spinach did not differ significantly by distribution channel. The data obtained in this study, as well as previously published data on the growth of (i) individual bacterial strains (i.e., strain-level growth) and (ii) the overall bacterial population on baby spinach (i.e., population-level growth), were used to fit models. Specifically, GP models were fit to population-level growth data only, while M models were fit to strain-level and population-level growth data. The RMSE values for the M models (i.e., 0.72, 0.77 and 1.09 log10 CFU/g, for three M models assessed here) and GP models (i.e., 0.68 and 0.81 log10 CFU/g, for the two GP models assessed here) are similar, which suggests that both M and GP models show comparable accuracy at predicting bacterial growth on spinach.

Neuroanatomical prediction of individual anxiety problems level using machine learning models: A population-based cohort study of young adults

Anxiety, a mental state in healthy individuals, is characterized by apprehension of potential future threats. Though the neurobiological basis of anxiety has been investigated widely in the clinical populations, the underly mechanism of neuroanatomical correlates with anxiety level in healthy young adults is still unclear. In this study, 1080 young adults were enrolled from the Human Connectome Project Young Adult dataset, and machine learning-based elastic net regression models with cross validation, together with linear mix effects (LME) models were adopted to investigate whether the neuroanatomical profiles of structural magnetic resonance imaging indicators associated with anxiety level in healthy young adults. We found multi-region neuroanatomical profiles predicted anxiety problems level and it was still robust in an out-of-sample. The neuroanatomical profiles had widespread brain nodes, including the dorsal lateral prefrontal cortex, supramarginal gyrus, and entorhinal cortex, which implicated in the default mode network and frontoparietal network. This finding was further supported by LME models, which showed significant univariate associations between brain nodes with anxiety. In sum, it's a large sample size study with multivariate analysis methodology to provide evidence that individual anxiety problems level can be predicted by machine learning-based models in healthy young adults. The neuroanatomical signature including hub nodes involved theoretically relevant brain networks robustly predicts anxiety, which could aid the assessment of potential high-risk of anxiety individuals.

The Transition From Homogeneous to Heterogeneous Machine Learning in Neuropsychiatric Research

Despite the advantage of neuroimaging-based machine learning (ML) models as pivotal tools for investigating brain-behavior relationships in neuropsychiatric studies, these data-driven predictive approaches have yet to yield substantial, clinically actionable insights for mental health care. A notable impediment lies in the inadequate accommodation of most ML research to the natural heterogeneity within large samples. Although commonly thought of as individual-level analyses, many ML algorithms are unimodal and homogeneous and thus incapable of capturing the potentially heterogeneous relationships between biology and psychopathology. We review the current landscape of computational research targeting population heterogeneity and argue that there is a need to expand from brain subtyping and behavioral phenotyping to analyses that focus on heterogeneity at the relational level. To this end, we review and suggest several existing ML models with the capacity to discern how external environmental and sociodemographic factors moderate the brain-behavior mapping function in a data-driven fashion. These heterogeneous ML models hold promise for enhancing the discovery of individualized brain-behavior associations and advancing precision psychiatry.

How measurement noise limits the accuracy of brain-behaviour predictions

Major efforts in human neuroimaging strive to understand individual differences and find biomarkers for clinical applications by predicting behavioural phenotypes from brain imaging data. To identify generalisable and replicable brain-behaviour prediction models, sufficient measurement reliability is essential. However, the selection of prediction targets is predominantly guided by scientific interest or data availability rather than psychometric considerations. Here, we demonstrate the impact of low reliability in behavioural phenotypes on out-of-sample prediction performance. Using simulated and empirical data from four large-scale datasets, we find that reliability levels common across many phenotypes can markedly limit the ability to link brain and behaviour. Next, using 5000 participants from the UK Biobank, we show that only highly reliable data can fully benefit from increasing sample sizes from hundreds to thousands of participants. Our findings highlight the importance of measurement reliability for identifying meaningful brain-behaviour associations from individual differences and underscore the need for greater emphasis on psychometrics in future research.

A data integration method for new advances in development cognitive neuroscience

Combining existing datasets to investigate key questions in developmental cognitive neuroscience brings exciting opportunities and unique challenges. However, many data pooling methods require identical or harmonized methodologies that are often not feasible. We propose Integrative Data Analysis (IDA) as a promising framework to advance developmental cognitive neuroscience with secondary data analysis. IDA serves to test hypotheses by combining data of the same construct from commensurate (but not identical) measures. To overcome idiosyncrasies of neuroimaging data, IDA explicitly evaluates if measures across studies assess the same construct. Moreover, IDA allows investigators to examine meaningful individual variability by de-confounding source-specific differences. To demonstrate IDA's potential, we explain foundational concepts, outline necessary steps, and apply IDA to volumetric measures of hippocampal subfields from 443 4- to 17-year-olds across three independent studies. We identified commensurate measures of Cornu Ammonis (CA) 1, dentate gyrus (DG)/CA3, and Subiculum (Sub). Model testing supported use of IDA to create IDA factor scores. We found age-related differences in DG/CA3, not but CA1 and Sub volume in the integrated dataset. By successfully demonstrating IDA, our hope is that future innovations come from the combination of existing neuroimaging data to create representative integrated samples when testing critical developmental questions.

Predictive and Explainable Artificial Intelligence for Neuroimaging Applications

BACKGROUND

The aim of this review is to highlight the new advance of predictive and explainable artificial intelligence for neuroimaging applications.

METHODS

Data came from 30 original studies in PubMed with the following search terms: "neuroimaging" (title) together with "machine learning" (title) or "deep learning" (title). The 30 original studies were eligible according to the following criteria: the participants with the dependent variable of brain image or associated disease; the interventions/comparisons of artificial intelligence; the outcomes of accuracy, the area under the curve (AUC), and/or variable importance; the publication year of 2019 or later; and the publication language of English.

RESULTS

The performance outcomes reported were within 58-96 for accuracy (%), 66-97 for sensitivity (%), 76-98 for specificity (%), and 70-98 for the AUC (%). The support vector machine and the convolutional neural network registered the best performance (AUC 98%) for the classifications of low- vs. high-grade glioma and brain conditions, respectively. Likewise, the random forest delivered the best performance (root mean square error 1) for the regression of brain conditions. The following factors were discovered to be major predictors of brain image or associated disease: (demographic) age, education, sex; (health-related) alpha desynchronization, Alzheimer's disease stage, CD4, depression, distress, mild behavioral impairment, RNA sequencing; (neuroimaging) abnormal amyloid-β, amplitude of low-frequency fluctuation, cortical thickness, functional connectivity, fractal dimension measure, gray matter volume, left amygdala activity, left hippocampal volume, plasma neurofilament light, right cerebellum, regional homogeneity, right middle occipital gyrus, surface area, sub-cortical volume.

CONCLUSION

Predictive and explainable artificial intelligence provide an effective, non-invasive decision support system for neuroimaging applications.

Network-level enrichment provides a framework for biological interpretation of machine learning results

Machine learning algorithms are increasingly being utilized to identify brain connectivity biomarkers linked to behavioral and clinical outcomes. However, research often prioritizes prediction accuracy at the expense of biological interpretability, and inconsistent implementation of ML methods may hinder model accuracy. To address this, our paper introduces a network-level enrichment approach, which integrates brain system organization in the context of connectome-wide statistical analysis to reveal network-level links between brain connectivity and behavior. To demonstrate the efficacy of this approach, we used linear support vector regression (LSVR) models to examine the relationship between resting-state functional connectivity networks and chronological age. We compared network-level associations based on raw LSVR weights to those produced from the forward and inverse models. Results indicated that not accounting for shared family variance inflated prediction performance, the k-best feature selection via Pearson correlation reduced accuracy and reliability, and raw LSVR model weights produced network-level associations that deviated from the significant brain systems identified by forward and inverse models. Our findings offer crucial insights for applying machine learning to neuroimaging data, emphasizing the value of network enrichment for biological interpretation.

TractoSCR: a novel supervised contrastive regression framework for prediction of neurocognitive measures using multi-site harmonized diffusion MRI tractography

Neuroimaging-based prediction of neurocognitive measures is valuable for studying how the brain's structure relates to cognitive function. However, the accuracy of prediction using popular linear regression models is relatively low. We propose a novel deep regression method, namely TractoSCR, that allows full supervision for contrastive learning in regression tasks using diffusion MRI tractography. TractoSCR performs supervised contrastive learning by using the absolute difference between continuous regression labels (i.e., neurocognitive scores) to determine positive and negative pairs. We apply TractoSCR to analyze a large-scale dataset including multi-site harmonized diffusion MRI and neurocognitive data from 8,735 participants in the Adolescent Brain Cognitive Development (ABCD) Study. We extract white matter microstructural measures using a fine parcellation of white matter tractography into fiber clusters. Using these measures, we predict three scores related to domains of higher-order cognition (general cognitive ability, executive function, and learning/memory). To identify important fiber clusters for prediction of these neurocognitive scores, we propose a permutation feature importance method for high-dimensional data. We find that TractoSCR obtains significantly higher accuracy of neurocognitive score prediction compared to other state-of-the-art methods. We find that the most predictive fiber clusters are predominantly located within the superficial white matter and projection tracts, particularly the superficial frontal white matter and striato-frontal connections. Overall, our results demonstrate the utility of contrastive representation learning methods for regression, and in particular for improving neuroimaging-based prediction of higher-order cognitive abilities. Our code will be available at: https://github.com/SlicerDMRI/TractoSCR.

Stable multivariate lesion symptom mapping

Multivariate lesion-symptom mapping (MLSM) considers lesion information across the entire brain to predict impairments. The strength of this approach is also its weakness-considering many brain features together synergistically can uncover complex brain-behavior relationships but exposes a high-dimensional feature space that a model is expected to learn. Successfully distinguishing between features in this landscape can be difficult for models, particularly in the presence of irrelevant or redundant features. Here, we propose stable multivariate lesion-symptom mapping (sMLSM), which integrates the identification of reliable features with stability selection into conventional MLSM and describe our open-source MATLAB implementation. Usage is showcased with our publicly available dataset of chronic stroke survivors (N=167) and further validated in our independent public acute stroke dataset (N = 1106). We demonstrate that sMLSM eliminates inconsistent features highlighted by MLSM, reduces variation in feature weights, enables the model to learn more complex patterns of brain damage, and improves model accuracy for predicting aphasia severity in a way that tends to be robust regarding the choice of parameters for identifying reliable features. Critically, sMLSM more consistently outperforms predictions based on lesion size alone. This advantage is evident starting at modest sample sizes (N>75). Spatial distribution of feature importance is different in sMLSM, which highlights the features identified by univariate lesion symptom mapping while also implicating select regions emphasized by MLSM. Beyond improved prediction accuracy, sMLSM can offer deeper insight into reliable biomarkers of impairment, informing our understanding of neurobiology.

Observing the suppression of individual aversive memories from conscious awareness

When reminded of an unpleasant experience, people often try to exclude the unwanted memory from awareness, a process known as retrieval suppression. Here we used multivariate decoding (MVPA) and representational similarity analyses on EEG data to track how suppression unfolds in time and to reveal its impact on item-specific cortical patterns. We presented reminders to aversive scenes and asked people to either suppress or to retrieve the scene. During suppression, mid-frontal theta power within the first 500 ms distinguished suppression from passive viewing of the reminder, indicating that suppression rapidly recruited control. During retrieval, we could discern EEG cortical patterns relating to individual memories-initially, based on theta-driven visual perception of the reminders (0 to 500 ms) and later, based on alpha-driven reinstatement of the aversive scene (500 to 3000 ms). Critically, suppressing retrieval weakened (during 360 to 600 ms) and eventually abolished item-specific cortical patterns, a robust effect that persisted until the reminder disappeared (780 to 3000 ms). Representational similarity analyses provided converging evidence that retrieval suppression weakened the representation of target scenes during the 500 to 3000 ms reinstatement window. Together, rapid top-down control during retrieval suppression abolished cortical patterns of individual memories, and precipitated later forgetting. These findings reveal a precise chronometry on the voluntary suppression of individual memories.

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.

Sex classification from functional brain connectivity: Generalization to multiple datasets

Machine learning (ML) approaches are increasingly being applied to neuroimaging data. Studies in neuroscience typically have to rely on a limited set of training data which may impair the generalizability of ML models. However, it is still unclear which kind of training sample is best suited to optimize generalization performance. In the present study, we systematically investigated the generalization performance of sex classification models trained on the parcelwise connectivity profile of either single samples or compound samples of two different sizes. Generalization performance was quantified in terms of mean across-sample classification accuracy and spatial consistency of accurately classifying parcels. Our results indicate that the generalization performance of parcelwise classifiers (pwCs) trained on single dataset samples is dependent on the specific test samples. Certain datasets seem to "match" in the sense that classifiers trained on a sample from one dataset achieved a high accuracy when tested on the respected other one and vice versa. The pwCs trained on the compound samples demonstrated overall highest generalization performance for all test samples, including one derived from a dataset not included in building the training samples. Thus, our results indicate that both a large sample size and a heterogeneous data composition of a training sample have a central role in achieving generalizable results.

Sex classification from functional brain connectivity: Generalization to multiple datasets Generalizability of sex classifiers

Machine learning (ML) approaches are increasingly being applied to neuroimaging data. Studies in neuroscience typically have to rely on a limited set of training data which may impair the generalizability of ML models. However, it is still unclear which kind of training sample is best suited to optimize generalization performance. In the present study, we systematically investigated the generalization performance of sex classification models trained on the parcelwise connectivity profile of either single samples or a compound sample containing data from four different datasets. Generalization performance was quantified in terms of mean across-sample classification accuracy and spatial consistency of accurately classifying parcels. Our results indicate that generalization performance of pwCs trained on single dataset samples is dependent on the specific test samples. Certain datasets seem to "match" in the sense that classifiers trained on a sample from one dataset achieved a high accuracy when tested on the respected other one and vice versa. The pwC trained on the compound sample demonstrated overall highest generalization performance for all test samples, including one derived from a dataset not included in building the training samples. Thus, our results indicate that a big and heterogenous training sample comprising data of multiple datasets is best suited to achieve generalizable results.

Navigating Neural Landscapes: A Comprehensive Review of Magnetic Resonance Imaging (MRI) and Magnetic Resonance Spectroscopy (MRS) Applications in Epilepsy

This review comprehensively explores the evolving role of neuroimaging, specifically magnetic resonance imaging (MRI) and magnetic resonance spectroscopy (MRS), in epilepsy research and clinical practice. Beginning with a concise overview of epilepsy, the discussion emphasizes the crucial importance of neuroimaging in diagnosing and managing this complex neurological disorder. The review delves into the applications of advanced MRI techniques, including high-field MRI, resting-state fMRI, and connectomics, highlighting their impact on refining our understanding of epilepsy's structural and functional dimensions. Additionally, it examines the integration of machine learning in the analysis of intricate neuroimaging data. Moving to the clinical domain, the review outlines the utility of neuroimaging in pre-surgical evaluations and the monitoring of treatment responses and disease progression. Despite significant strides, challenges and limitations are discussed in the routine clinical incorporation of neuroimaging. The review explores promising developments in MRI and MRS technology, potential advancements in imaging biomarkers, and the implications for personalized medicine in epilepsy management. The conclusion underscores the transformative potential of neuroimaging and advocates for continued exploration, collaboration, and technological innovation to propel the field toward a future where tailored, effective interventions improve outcomes for individuals with epilepsy.

Slower speed of blood pressure recovery after standing is associated with accelerated brain aging: Evidence from The Irish Longitudinal Study on Ageing (TILDA)

Background

Impaired recovery of blood pressure (BP) in response to standing up is a prevalent condition in older individuals. We evaluated the relationship between the early recovery of hemodynamic responses to standing and brain health in adults over 50.

Methods

Participants from The Irish Longitudinal Study on Ageing (TILDA) (n=411; age 67.6 ± 7.3 years; 53.4 % women) performed an active stand challenge while blood pressure and heart rate were continuously monitored. The recovery of these parameters was determined as the slope of the BP and HR response, following the initial drop/rise after standing. We have previously reported a novel and validated measure of brain ageing using MRI data, which measures the difference between biological brain age and chronological age, providing a brain-predicted age difference (brainPAD) score.

Results

Slower recovery of systolic and diastolic BP was found to be significantly associated with higher brainPAD scores (i.e., biologically older brains), where a one-year increase in brainPAD was associated with a decrease of 0.02 mmHg/s and 0.01 mmHg/s in systolic and diastolic BP recovery, respectively, after standing. Heart rate (HR) recovery was not significantly associated with brainPAD score.

Conclusion

These results demonstrate that slower systolic and diastolic BP recovery in the early phase after standing is associated with accelerated brain aging in older individuals. This suggests that the BP response to standing, measured using beat-to-beat monitoring, has the potential to be used as a marker of accelerated brain aging, relying on a simple procedure and devices that are easily accessible.

Prediction of cognitive performance differences in older age from multimodal neuroimaging data

Differences in brain structure and functional and structural network architecture have been found to partly explain cognitive performance differences in older ages. Thus, they may serve as potential markers for these differences. Initial unimodal studies, however, have reported mixed prediction results of selective cognitive variables based on these brain features using machine learning (ML). Thus, the aim of the current study was to investigate the general validity of cognitive performance prediction from imaging data in healthy older adults. In particular, the focus was with examining whether (1) multimodal information, i.e., region-wise grey matter volume (GMV), resting-state functional connectivity (RSFC), and structural connectivity (SC) estimates, may improve predictability of cognitive targets, (2) predictability differences arise for global cognition and distinct cognitive profiles, and (3) results generalize across different ML approaches in 594 healthy older adults (age range: 55-85 years) from the 1000BRAINS study. Prediction potential was examined for each modality and all multimodal combinations, with and without confound (i.e., age, education, and sex) regression across different analytic options, i.e., variations in algorithms, feature sets, and multimodal approaches (i.e., concatenation vs. stacking). Results showed that prediction performance differed considerably between deconfounding strategies. In the absence of demographic confounder control, successful prediction of cognitive performance could be observed across analytic choices. Combination of different modalities tended to marginally improve predictability of cognitive performance compared to single modalities. Importantly, all previously described effects vanished in the strict confounder control condition. Despite a small trend for a multimodal benefit, developing a biomarker for cognitive aging remains challenging.

Neuroimaging and artificial intelligence for assessment of chronic painful temporomandibular disorders-a comprehensive review

Chronic Painful Temporomandibular Disorders (TMD) are challenging to diagnose and manage due to their complexity and lack of understanding of brain mechanism. In the past few decades' neural mechanisms of pain regulation and perception have been clarified by neuroimaging research. Advances in the neuroimaging have bridged the gap between brain activity and the subjective experience of pain. Neuroimaging has also made strides toward separating the neural mechanisms underlying the chronic painful TMD. Recently, Artificial Intelligence (AI) is transforming various sectors by automating tasks that previously required humans' intelligence to complete. AI has started to contribute to the recognition, assessment, and understanding of painful TMD. The application of AI and neuroimaging in understanding the pathophysiology and diagnosis of chronic painful TMD are still in its early stages. The objective of the present review is to identify the contemporary neuroimaging approaches such as structural, functional, and molecular techniques that have been used to investigate the brain of chronic painful TMD individuals. Furthermore, this review guides practitioners on relevant aspects of AI and how AI and neuroimaging methods can revolutionize our understanding on the mechanisms of painful TMD and aid in both diagnosis and management to enhance patient outcomes.

White matter predictors of PTSD: Testing different machine learning models in a sample of Black American women

BACKGROUND

Machine learning neuroimaging studies of posttraumatic stress disorder (PTSD) show promise for identifying neurobiological signatures of PTSD. However, studies to date, have largely evaluated a single machine learning approach, and few studies have examined white matter microstructure as a predictor of PTSD. Further, individuals from minoritized racial groups, specifically, Black individuals, who experience disproportionate trauma frequency, and have relatively higher rates of PTSD, have been underrepresented in these studies. We used four different machine learning models to test white matter microstructure classifiers of PTSD in a sample of trauma-exposed Black American women with and without PTSD.

METHOD

Participants included 45 Black women with PTSD and 89 trauma-exposed controls recruited from an ongoing trauma study. Current PTSD presence was estimated using the Clinician-Administered PTSD Scale. Average fractional anisotropy of 53 white matter tracts served as input features. Additional exploratory analysis incorporated estimates of interpersonal and structural racism exposure. Classification models included linear support vector machine, radial basis function support vector machine, multilayer perceptron, and random forest.

RESULTS

Performance varied notably between models. With white matter features along, linear support vector machine demonstrated the best model fit and reached an average AUC = 0.643. Inclusion of estimates of exposure to racism increased linear support vector machine performance (AUC = 0.808).

CONCLUSIONS

White matter microstructure had limited ability to predict PTSD presence in this sample. These results may indicate that the relationship between white matter microstructure and PTSD may be nuanced across race and gender spectrums.

Gaussian Process-based prediction of memory performance and biomarker status in ageing and Alzheimer's disease-A systematic model evaluation

Neuroimaging markers based on Magnetic Resonance Imaging (MRI) combined with various other measures (such as genetic covariates, biomarkers, vascular risk factors, neuropsychological tests etc.) might provide useful predictions of clinical outcomes during the progression towards Alzheimer's disease (AD). The use of multiple features in predictive frameworks for clinical outcomes has become increasingly prevalent in AD research. However, many studies do not focus on systematically and accurately evaluating combinations of multiple input features. Hence, the aim of the present work is to explore and assess optimal combinations of various features for MR-based prediction of (1) cognitive status and (2) biomarker positivity with a multi-kernel learning Gaussian process framework. The explored features and parameters included (A) combinations of brain tissues, modulation, smoothing, and image resolution; (B) incorporating demographics & clinical covariates; (C) the impact of the size of the training data set; (D) the influence of dimensionality reduction and the choice of kernel types. The approach was tested in a large German cohort including 959 subjects from the multicentric longitudinal study of cognitive impairment and dementia (DELCODE). Our evaluation suggests the best prediction of memory performance was obtained for a combination of neuroimaging markers, demographics, genetic information (ApoE4) and CSF biomarkers explaining 57% of outcome variance in out-of-sample predictions. The highest performance for Aβ42/40 status classification was achieved for a combination of demographics, ApoE4, and a memory score while usage of structural MRI further improved the classification of individual patient's pTau status.

Stress-induced brain responses are associated with BMI in women

Overweight and obesity are associated with altered stress reactivity and increased inflammation. However, it is not known whether stress-induced changes in brain function scale with BMI and if such associations are driven by peripheral cytokines. Here, we investigate multimodal stress responses in a large transdiagnostic sample using predictive modeling based on spatio-temporal profiles of stress-induced changes in activation and functional connectivity. BMI is associated with increased brain responses as well as greater negative affect after stress and individual response profiles are associated with BMI in females (pperm < 0.001), but not males. Although stress-induced changes reflecting BMI are associated with baseline cortisol, there is no robust association with peripheral cytokines. To conclude, alterations in body weight and energy metabolism might scale acute brain responses to stress more strongly in females compared to males, echoing observational studies. Our findings highlight sex-dependent associations of stress with differences in endocrine markers, largely independent of peripheral inflammation.

Accurate sex prediction of cisgender and transgender individuals without brain size bias

The increasing use of machine learning approaches on neuroimaging data comes with the important concern of confounding variables which might lead to biased predictions and in turn spurious conclusions about the relationship between the features and the target. A prominent example is the brain size difference between women and men. This difference in total intracranial volume (TIV) can cause bias when employing machine learning approaches for the investigation of sex differences in brain morphology. A TIV-biased model will not capture qualitative sex differences in brain organization but rather learn to classify an individual's sex based on brain size differences, thus leading to spurious and misleading conclusions, for example when comparing brain morphology between cisgender- and transgender individuals. In this study, TIV bias in sex classification models applied to cis- and transgender individuals was systematically investigated by controlling for TIV either through featurewise confound removal or by matching the training samples for TIV. Our results provide strong evidence that models not biased by TIV can classify the sex of both cis- and transgender individuals with high accuracy, highlighting the importance of appropriate modeling to avoid bias in automated decision making.

Machine-learning for the prediction of one-year seizure recurrence based on routine electroencephalography

Predicting seizure recurrence risk is critical to the diagnosis and management of epilepsy. Routine electroencephalography (EEG) is a cornerstone of the estimation of seizure recurrence risk. However, EEG interpretation relies on the visual identification of interictal epileptiform discharges (IEDs) by neurologists, with limited sensitivity. Automated processing of EEG could increase its diagnostic yield and accessibility. The main objective was to develop a prediction model based on automated EEG processing to predict one-year seizure recurrence in patients undergoing routine EEG. We retrospectively selected a consecutive cohort of 517 patients undergoing routine EEG at our institution (training set) and a separate, temporally shifted cohort of 261 patients (testing set). We developed an automated processing pipeline to extract linear and non-linear features from the EEGs. We trained machine learning algorithms on multichannel EEG segments to predict one-year seizure recurrence. We evaluated the impact of IEDs and clinical confounders on performances and validated the performances on the testing set. The receiver operating characteristic area-under-the-curve for seizure recurrence after EEG in the testing set was 0.63 (95% CI 0.55-0.71). Predictions were still significantly above chance in EEGs with no IEDs. Our findings suggest that there are changes other than IEDs in the EEG signal embodying seizure propensity.

Decoding individual differences in self-prioritization from the resting-state functional connectome

Although the self has traditionally been viewed as a higher-order mental function by most theoretical frameworks, recent research advocates a fundamental self hypothesis, viewing the self as a baseline function of the brain embedded within its spontaneous activities, which dynamically regulates cognitive processing and subsequently guides behavior. Understanding this fundamental self hypothesis can reveal where self-biased behaviors emerge and to what extent brain signals at rest can predict such biased behaviors. To test this hypothesis, we investigated the association between spontaneous neural connectivity and robust self-bias in a perceptual matching task using resting-state functional magnetic resonance imaging (fMRI) in 348 young participants. By decoding whole-brain connectivity patterns, the support vector regression model produced the best predictions of the magnitude of self-bias in behavior, which was evaluated via a nested cross-validation procedure. The out-of-sample generalizability was further authenticated using an external dataset of older adults. The functional connectivity results demonstrated that self-biased behavior was associated with distinct connections between the default mode, cognitive control, and salience networks. Consensus network and computational lesion analyses further revealed contributing regions distributed across six networks, extending to additional nodes, such as the thalamus, whose role in self-related processing remained unclear. These results provide evidence that self-biased behavior derives from spontaneous neural connectivity, supporting the fundamental self hypothesis. Thus, we propose an integrated neural network model of this fundamental self that synthesizes previous theoretical models and portrays the brain mechanisms by which the self emerges at rest internally and regulates responses to the external environment.

The confound of hemodynamic response function variability in human resting-state functional MRI studies

Functional magnetic resonance imaging (fMRI) is an indirect measure of neural activity with the hemodynamic response function (HRF) coupling it with unmeasured neural activity. The HRF, modulated by several non-neural factors, is variable across brain regions, individuals and populations. Yet, a majority of human resting-state fMRI connectivity studies continue to assume a non-variable HRF. In this article, with supportive prior evidence, we argue that HRF variability cannot be ignored as it substantially confounds within-subject connectivity estimates and between-subjects connectivity group differences. We also discuss its clinical relevance with connectivity impairments confounded by HRF aberrations in several disorders. We present limited data on HRF differences between women and men, which resulted in a 15.4% median error in functional connectivity estimates in a group-level comparison. We also discuss the implications of HRF variability for fMRI studies in the spinal cord. There is a need for more dialogue within the community on the HRF confound, and we hope that our article is a catalyst in the process.

Using histogram analysis of the intrinsic brain activity mapping to identify essential tremor

Background

Essential tremor (ET) is one of the most common movement disorders. Histogram analysis based on brain intrinsic activity imaging is a promising way to identify ET patients from healthy controls (HCs) and further explore the spontaneous brain activity change mechanisms and build the potential diagnostic biomarker in ET patients.

Methods

The histogram features based on the Resting-state functional magnetic resonance imaging (Rs-fMRI) data were extracted from 133 ET patients and 135 well-matched HCs as the input features. Then, a two-sample t-test, the mutual information, and the least absolute shrinkage and selection operator methods were applied to reduce the feature dimensionality. Support vector machine (SVM), logistic regression (LR), random forest (RF), and k-nearest neighbor (KNN) were used to differentiate ET and HCs, and classification performance of the established models was evaluated by the mean area under the curve (AUC). Moreover, correlation analysis was carried out between the selected histogram features and clinical tremor characteristics.

Results

Each classifier achieved a good classification performance in training and testing sets. The mean accuracy and area under the curve (AUC) of SVM, LR, RF, and KNN in the testing set were 92.62%, 0.948; 92.01%, 0.942; 93.88%, 0.941; and 92.27%, 0.939, respectively. The most power-discriminative features were mainly located in the cerebello-thalamo-motor and non-motor cortical pathways. Correlation analysis showed that there were two histogram features negatively and one positively correlated with tremor severity.

Conclusion

Our findings demonstrated that the histogram analysis of the amplitude of low-frequency fluctuation (ALFF) images with multiple machine learning algorithms could identify ET patients from HCs and help to understand the spontaneous brain activity pathogenesis mechanisms in ET patients.

Two sides of the same coin: distinct neuroanatomical patterns predict crystallized and fluid intelligence in adults

Background

Crystallized intelligence (Gc) and fluid intelligence (Gf) are regarded as distinct intelligence components that statistically correlate with each other. However, the distinct neuroanatomical signatures of Gc and Gf in adults remain contentious.

Methods

Machine learning cross-validated elastic net regression models were performed on the Human Connectome Project Young Adult dataset (N = 1089) to characterize the neuroanatomical patterns of structural magnetic resonance imaging variables that are associated with Gc and Gf. The observed relationships were further examined by linear mixed-effects models. Finally, intraclass correlations were computed to examine the similarity of the neuroanatomical correlates between Gc and Gf.

Results

The results revealed distinct multi-region neuroanatomical patterns predicted Gc and Gf, respectively, which were robust in a held-out test set (R² = 2.40, 1.97%, respectively). The relationship of these regions with Gc and Gf was further supported by the univariate linear mixed effects models. Besides that, Gc and Gf displayed poor neuroanatomical similarity.

Conclusion

These findings provided evidence that distinct machine learning-derived neuroanatomical patterns could predict Gc and Gf in healthy adults, highlighting differential neuroanatomical signatures of different aspects of intelligence.

Automated identification of piglet brain tissue from MRI images using Region-based Convolutional Neural Networks

Magnetic resonance imaging is an important tool for characterizing volumetric changes of the piglet brain during development. Typically, an early step of an imaging analysis pipeline is brain extraction, or skull stripping. Brain extractions are usually performed manually; however, this approach is time-intensive and can lead to variation between brain extractions when multiple raters are used. Automated brain extractions are important for reducing the time required for analyses and improving the uniformity of the extractions. Here we demonstrate the use of Mask R-CNN, a Region-based Convolutional Neural Network (R-CNN), for automated brain extractions of piglet brains. We validate our approach using Nested Cross-Validation on six sets of training/validation data drawn from 32 pigs. Visual inspection of the extractions shows acceptable accuracy, Dice coefficients are in the range of 0.95-0.97, and Hausdorff Distance values in the range of 4.1-8.3 voxels. These results demonstrate that R-CNNs provide a viable tool for skull stripping of piglet brains.

MLATE: Machine learning for predicting cell behavior on cardiac tissue engineering scaffolds

Cardiovascular disease is one of the leading causes of mortality worldwide and is responsible for millions of deaths annually. One of the most promising approaches to deal with this problem, which has spread recently, is cardiac tissue engineering (CTE). Many researchers have tried developing scaffolds with different materials, cell lines, and fabrication methods to help regenerate heart tissue. Machine learning (ML) is one of the hottest topics in science and technology, revolutionizing many fields and changing our perspective on solving problems. As a result of using ML, some scientific issues have been resolved, including protein-folding, a challenging problem in biology that remained unsolved for 50 years. However, it is not well addressed in tissue engineering. An AI-based software was developed by our group called MLATE (Machine Learning Applications in Tissue Engineering) to tackle tissue engineering challenges, which highly depend on conducting costly and time-consuming experiments. For the first time, to the best of our knowledge, a CTE scaffold dataset was created by collecting specifications from the literature, including different materials, cell lines, and fabrication methods commonly used in CTE scaffold development. These specifications were used as variables in the study. Then, the CTE scaffolds were rated based on cell behaviors such as cell viability, growth, proliferation, and differentiation on the scaffold on a scale of 0-3. These ratings were considered a function of the variables in the gathered dataset. It should be stated that this study was merely based on information available in the literature. Then, twenty-eight ML algorithms were applied to determine the most effective one for predicting cell behavior on CTE scaffolds fabricated by different materials, compositions, and methods. The results indicated the high performance of XGBoost with an accuracy of 87%. Also, by implementing ensemble learning algorithms and using five algorithms with the best performance, an accuracy of 93% with the AdaBoost Classifier and Voting Classifier was achieved. Finally, the open-source software developed in this study was made available for everyone by publishing the best model along with a step-by-step guide to using it online at: https://github.com/saeedrafieyan/MLATE.

Artificial intelligence for diffusion MRI-based tissue microstructure estimation in the human brain: an overview

Artificial intelligence (AI) has made significant advances in the field of diffusion magnetic resonance imaging (dMRI) and other neuroimaging modalities. These techniques have been applied to various areas such as image reconstruction, denoising, detecting and removing artifacts, segmentation, tissue microstructure modeling, brain connectivity analysis, and diagnosis support. State-of-the-art AI algorithms have the potential to leverage optimization techniques in dMRI to advance sensitivity and inference through biophysical models. While the use of AI in brain microstructures has the potential to revolutionize the way we study the brain and understand brain disorders, we need to be aware of the pitfalls and emerging best practices that can further advance this field. Additionally, since dMRI scans rely on sampling of the q-space geometry, it leaves room for creativity in data engineering in such a way that it maximizes the prior inference. Utilization of the inherent geometry has been shown to improve general inference quality and might be more reliable in identifying pathological differences. We acknowledge and classify AI-based approaches for dMRI using these unifying characteristics. This article also highlighted and reviewed general practices and pitfalls involving tissue microstructure estimation through data-driven techniques and provided directions for building on them.

One Size Does Not Fit All: Methodological Considerations for Brain-Based Predictive Modeling in Psychiatry

Psychiatric illnesses are heterogeneous in nature. No illness manifests in the same way across individuals, and no two patients with a shared diagnosis exhibit identical symptom profiles. Over the last several decades, group-level analyses of in vivo neuroimaging data have led to fundamental advances in our understanding of the neurobiology of psychiatric illnesses. More recently, access to computational resources and large, publicly available datasets alongside the rise of predictive modeling and precision medicine approaches have facilitated the study of psychiatric illnesses at an individual level. Data-driven machine learning analyses can be applied to identify disease-relevant biological subtypes, predict individual symptom profiles, and recommend personalized therapeutic interventions. However, when developing these predictive models, methodological choices must be carefully considered to ensure accurate, robust, and interpretable results. Choices pertaining to algorithms, neuroimaging modalities and states, data transformation, phenotypes, parcellations, sample sizes, and populations we are specifically studying can influence model performance. Here, we review applications of neuroimaging-based machine learning models to study psychiatric illnesses and discuss the effects of different methodological choices on model performance. An understanding of these effects is crucial for the proper implementation of predictive models in psychiatry and will facilitate more accurate diagnoses, prognoses, and therapeutics.

Polyneuro risk scores capture widely distributed connectivity patterns of cognition

Resting-state functional connectivity (RSFC) is a powerful tool for characterizing brain changes, but it has yet to reliably predict higher-order cognition. This may be attributed to small effect sizes of such brain-behavior relationships, which can lead to underpowered, variable results when utilizing typical sample sizes (N∼25). Inspired by techniques in genomics, we implement the polyneuro risk score (PNRS) framework - the application of multivariate techniques to RSFC data and validation in an independent sample. Utilizing the Adolescent Brain Cognitive Development® cohort split into two datasets, we explore the framework's ability to reliably capture brain-behavior relationships across 3 cognitive scores - general ability, executive function, learning & memory. The weight and significance of each connection is assessed in the first dataset, and a PNRS is calculated for each participant in the second. Results support the PNRS framework as a suitable methodology to inspect the distribution of connections contributing towards behavior, with explained variance ranging from 1.0 % to 21.4 %. For the outcomes assessed, the framework reveals globally distributed, rather than localized, patterns of predictive connections. Larger samples are likely necessary to systematically identify the specific connections contributing towards complex outcomes. The PNRS framework could be applied translationally to identify neurologically distinct subtypes of neurodevelopmental disorders.

Supervised Phenotype Discovery From Multimodal Brain Imaging

Data-driven discovery of image-derived phenotypes (IDPs) from large-scale multimodal brain imaging data has enormous potential for neuroscientific and clinical research by linking IDPs to subjects' demographic, behavioural, clinical and cognitive measures (i.e., non-imaging derived phenotypes or nIDPs). However, current approaches are primarily based on unsupervised approaches, without the use of information in nIDPs. In this paper, we proposed a semi-supervised, multimodal, and multi-task fusion approach, termed SuperBigFLICA, for IDP discovery, which simultaneously integrates information from multiple imaging modalities as well as multiple nIDPs. SuperBigFLICA is computationally efficient and largely avoids the need for parameter tuning. Using the UK Biobank brain imaging dataset with around 40,000 subjects and 47 modalities, along with more than 17,000 nIDPs, we showed that SuperBigFLICA enhances the prediction power of nIDPs, benchmarked against IDPs derived by conventional expert-knowledge and unsupervised-learning approaches (with average nIDP prediction accuracy improvements of up to 46%). It also enables the learning of generic imaging features that can predict new nIDPs. Further empirical analysis of the SuperBigFLICA algorithm demonstrates its robustness in different prediction tasks and the ability to derive biologically meaningful IDPs in predicting health outcomes and cognitive nIDPs, such as fluid intelligence and hypertension.

Machine learning approaches linking brain function to behavior in the ABCD STOP task

The stop-signal task (SST) is one of the most common fMRI tasks of response inhibition, and its performance measure, the stop-signal reaction-time (SSRT), is broadly used as a measure of cognitive control processes. The neurobiology underlying individual or clinical differences in response inhibition remain unclear, consistent with the general pattern of quite modest brain-behavior associations that have been recently reported in well-powered large-sample studies. Here, we investigated the potential of multivariate, machine learning (ML) methods to improve the estimation of individual differences in SSRT with multimodal structural and functional region of interest-level neuroimaging data from 9- to 11-year-olds children in the ABCD Study. Six ML algorithms were assessed across modalities and fMRI tasks. We verified that SST activation performed best in predicting SSRT among multiple modalities including morphological MRI (cortical surface area/thickness), diffusion tensor imaging, and fMRI task activations, and then showed that SST activation explained 12% of the variance in SSRT using cross-validation and out-of-sample lockbox data sets (n = 7298). Brain regions that were more active during the task and that showed more interindividual variation in activation were better at capturing individual differences in performance on the task, but this was only true for activations when successfully inhibiting. Cortical regions outperformed subcortical areas in explaining individual differences but the two hemispheres performed equally well. These results demonstrate that the detection of reproducible links between brain function and performance can be improved with multivariate approaches and give insight into a number of brain systems contributing to individual differences in this fundamental cognitive control process.

Evaluation of Risk of Bias in Neuroimaging-Based Artificial Intelligence Models for Psychiatric Diagnosis: A Systematic Review

IMPORTANCE

Neuroimaging-based artificial intelligence (AI) diagnostic models have proliferated in psychiatry. However, their clinical applicability and reporting quality (ie, feasibility) for clinical practice have not been systematically evaluated.

OBJECTIVE

To systematically assess the risk of bias (ROB) and reporting quality of neuroimaging-based AI models for psychiatric diagnosis.

EVIDENCE REVIEW

PubMed was searched for peer-reviewed, full-length articles published between January 1, 1990, and March 16, 2022. Studies aimed at developing or validating neuroimaging-based AI models for clinical diagnosis of psychiatric disorders were included. Reference lists were further searched for suitable original studies. Data extraction followed the CHARMS (Checklist for Critical Appraisal and Data Extraction for Systematic Reviews of Prediction Modeling Studies) and PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) guidelines. A closed-loop cross-sequential design was used for quality control. The PROBAST (Prediction Model Risk of Bias Assessment Tool) and modified CLEAR (Checklist for Evaluation of Image-Based Artificial Intelligence Reports) benchmarks were used to systematically evaluate ROB and reporting quality.

FINDINGS

A total of 517 studies presenting 555 AI models were included and evaluated. Of these models, 461 (83.1%; 95% CI, 80.0%-86.2%) were rated as having a high overall ROB based on the PROBAST. The ROB was particular high in the analysis domain, including inadequate sample size (398 of 555 models [71.7%; 95% CI, 68.0%-75.6%]), poor model performance examination (with 100% of models lacking calibration examination), and lack of handling data complexity (550 of 555 models [99.1%; 95% CI, 98.3%-99.9%]). None of the AI models was perceived to be applicable to clinical practices. Overall reporting completeness (ie, number of reported items/number of total items) for the AI models was 61.2% (95% CI, 60.6%-61.8%), and the completeness was poorest for the technical assessment domain with 39.9% (95% CI, 38.8%-41.1%).

CONCLUSIONS AND RELEVANCE

This systematic review found that the clinical applicability and feasibility of neuroimaging-based AI models for psychiatric diagnosis were challenged by a high ROB and poor reporting quality. Particularly in the analysis domain, ROB in AI diagnostic models should be addressed before clinical application.

Hemodynamic MRI parameters to predict asymptomatic unilateral carotid artery stenosis with random forest machine learning

BACKGROUND

Internal carotid artery stenosis (ICAS) can cause stroke and cognitive decline. Associated hemodynamic impairments, which are most pronounced within individual watershed areas (iWSA) between vascular territories, can be assessed with hemodynamic-oxygenation-sensitive MRI and may help to detect severely affected patients. We aimed to identify the most sensitive parameters and volumes of interest (VOI) to predict high-grade ICAS with random forest machine learning. We hypothesized an increased predictive ability considering iWSAs and a decreased cognitive performance in correctly classified patients.

MATERIALS AND METHODS

Twenty-four patients with asymptomatic, unilateral, high-grade carotid artery stenosis and 24 age-matched healthy controls underwent MRI comprising pseudo-continuous arterial spin labeling (pCASL), breath-holding functional MRI (BH-fMRI), dynamic susceptibility contrast (DSC), T2 and T2* mapping, MPRAGE and FLAIR. Quantitative maps of eight perfusion, oxygenation and microvascular parameters were obtained. Mean values of respective parameters within and outside of iWSAs split into gray (GM) and white matter (WM) were calculated for both hemispheres and for interhemispheric differences resulting in 96 features. Random forest classifiers were trained on whole GM/WM VOIs, VOIs considering iWSAs and with additional feature selection, respectively.

RESULTS

The most sensitive features in decreasing order were time-to-peak (TTP), cerebral blood flow (CBF) and cerebral vascular reactivity (CVR), all of these inside of iWSAs. Applying iWSAs combined with feature selection yielded significantly higher receiver operating characteristics areas under the curve (AUC) than whole GM/WM VOIs (AUC: 0.84 vs. 0.90, p = 0.039). Correctly predicted patients presented with worse cognitive performances than frequently misclassified patients (Trail-making-test B: 152.5s vs. 94.4s, p = 0.034).

CONCLUSION

Random forest classifiers trained on multiparametric MRI data allow identification of the most relevant parameters and VOIs to predict ICAS, which may improve personalized treatments.

Threats to external validity in the neuroprediction of substance use treatment outcomes

Background: Tools predicting individual relapse risk would invaluably inform clinical decision-making (e.g. level-of-care) in substance use treatment. Studies of neuroprediction - use of neuromarkers to predict individual outcomes - have the dual potential to create such tools and inform etiological models leading to new treatments. However, financial limitations, statistical power demands, and related factors encourage restrictive selection criteria, yielding samples that do not fully represent the target population. This problem may be further compounded by a lack of statistical optimism correction in neuroprediction research, resulting in predictive models that are overfit to already-restricted samples.Objectives: This systematic review aims to identify potential threats to external validity related to restrictive selection criteria and underutilization of optimism correction in the existing neuroprediction literature targeting substance use treatment outcomes.Methods: Sixty-seven studies of neuroprediction in substance use treatment were identified and details of sample selection criteria and statistical optimism correction were extracted.Results: Most publications were found to report restrictive selection criteria (e.g. excluding psychiatric (94% of publications) and substance use comorbidities (69% of publications)) that would rule-out a considerable portion of the treatment population. Furthermore, only 21% of publications reported optimism correction.Conclusion: Restrictive selection criteria and underutilization of optimism correction are common in the existing literature and may limit the generalizability of identified neural predictors to the target population whose treatment they would ultimately inform. Greater attention to the inclusivity and generalizability of addiction neuroprediction research, as well as new opportunities provided through open science initiatives, have the potential to address this issue.

Data management strategy for a collaborative research center

The importance of effective research data management (RDM) strategies to support the generation of Findable, Accessible, Interoperable, and Reusable (FAIR) neuroscience data grows with each advance in data acquisition techniques and research methods. To maximize the impact of diverse research strategies, multidisciplinary, large-scale neuroscience research consortia face a number of unsolved challenges in RDM. While open science principles are largely accepted, it is practically difficult for researchers to prioritize RDM over other pressing demands. The implementation of a coherent, executable RDM plan for consortia spanning animal, human, and clinical studies is becoming increasingly challenging. Here, we present an RDM strategy implemented for the Heidelberg Collaborative Research Consortium. Our consortium combines basic and clinical research in diverse populations (animals and humans) and produces highly heterogeneous and multimodal research data (e.g., neurophysiology, neuroimaging, genetics, behavior). We present a concrete strategy for initiating early-stage RDM and FAIR data generation for large-scale collaborative research consortia, with a focus on sustainable solutions that incentivize incremental RDM while respecting research-specific requirements.

Performance scaling for structural MRI surface parcellations: a machine learning analysis in the ABCD Study

The use of predefined parcellations on surface-based representations of the brain as a method for data reduction is common across neuroimaging studies. In particular, prediction-based studies typically employ parcellation-driven summaries of brain measures as input to predictive algorithms, but the choice of parcellation and its influence on performance is often ignored. Here we employed preprocessed structural magnetic resonance imaging (sMRI) data from the Adolescent Brain Cognitive Development Study® to examine the relationship between 220 parcellations and out-of-sample predictive performance across 45 phenotypic measures in a large sample of 9- to 10-year-old children (N = 9,432). Choice of machine learning (ML) pipeline and use of alternative multiple parcellation-based strategies were also assessed. Relative parcellation performance was dependent on the spatial resolution of the parcellation, with larger number of parcels (up to ~4,000) outperforming coarser parcellations, according to a power-law scaling of between 1/4 and 1/3. Performance was further influenced by the type of parcellation, ML pipeline, and general strategy, with existing literature-based parcellations, a support vector-based pipeline, and ensembling across multiple parcellations, respectively, as the highest performing. These findings highlight the choice of parcellation as an important influence on downstream predictive performance, showing in some cases that switching to a higher resolution parcellation can yield a relatively large boost to performance.

Predicting the severity of internet gaming disorder with resting-state brain features: A multi-voxel pattern analysis

BACKGROUND

Internet gaming disorder (IGD) has become a worldwide mental health concern; however, the neural mechanism underlying this disorder remains unclear. Multivoxel pattern analysis (MVPA), a newly developed data-driven approach, can be used to investigate the neural features of IGD based on massive neural data.

METHODS

Resting-state fMRI data from four hundred and two participants with varying levels of IGD severity were recruited. Regional homogeneity (ReHo) and the amplitude of low-frequency fluctuation (ALFF) were calculated and subsequently decoded by applying MVPA. The highly weighted regions in both predictive models were selected as regions of interest for further graph theory and Granger causality analysis (GCA) to explore how they affect IGD severity.

RESULTS

The results revealed that the neural patterns of ReHo and ALFF can independently and significantly predict IGD severity. The highly weighted regions that contributed to both predictive models were the right precentral gyrus and left postcentral gyrus. Moreover, topological properties of the right precentral gyrus were significantly correlated with IGD severity; further GCA revealed effective connectivity from the right precentral gyrus to left precentral gyrus and dorsal anterior cingulate cortex, both of which were significantly associated with IGD severity.

CONCLUSIONS

The present study demonstrated that IGD has distinctive neural patterns, and this pattern could be found by machine learning. In addition, the neural features in the right precentral gyrus play a key role in predicting IGD severity. The current study revealed the neural features of IGD and provided a potential target for IGD interventions using brain modulation.

Multimodal magnetic resonance neuroimaging measures characteristic of early cART-treated pediatric HIV: A feature selection approach

Children with perinatally acquired HIV (CPHIV) have poor cognitive outcomes despite early combination antiretroviral therapy (cART). While CPHIV-related brain alterations can be investigated separately using proton magnetic resonance spectroscopy (1 H-MRS), structural magnetic resonance imaging (sMRI), diffusion tensor imaging (DTI), and functional MRI (fMRI), a set of multimodal MRI measures characteristic of children on cART has not been previously identified. We used the embedded feature selection of a logistic elastic-net (EN) regularization to select neuroimaging measures that distinguish CPHIV from controls and measured their classification performance via the area under the receiver operating characteristic curve (AUC) using repeated cross validation. We also wished to establish whether combining MRI modalities improved the models. In single modality analysis, sMRI volumes performed best followed by DTI, whereas individual EN models on spectroscopic, gyrification, and cortical thickness measures showed no class discrimination capability. Adding DTI and 1 H-MRS in basal measures to sMRI volumes produced the highest classification performancevalidation accuracy = 85 % AUC = 0.80 . The best multimodal MRI set consisted of 22 DTI and sMRI volume features, which included reduced volumes of the bilateral globus pallidus and amygdala, as well as increased mean diffusivity (MD) and radial diffusivity (RD) in the right corticospinal tract in cART-treated CPHIV. Consistent with previous studies of CPHIV, select subcortical volumes obtained from sMRI provide reasonable discrimination between CPHIV and controls. This may give insight into neuroimaging measures that are relevant in understanding the effects of HIV on the brain, thereby providing a starting point for evaluating their link with cognitive performance in CPHIV.

Large-scale comparison of machine learning algorithms for target prediction of natural products

Natural products (NPs) and their derivatives are important resources for drug discovery. There are many in silico target prediction methods that have been reported, however, very few of them distinguish NPs from synthetic molecules. Considering the fact that NPs and synthetic molecules are very different in many characteristics, it is necessary to build specific target prediction models of NPs. Therefore, we collected the activity data of NPs and their derivatives from the public databases and constructed four datasets, including the NP dataset, the NPs and its first-class derivatives dataset, the NPs and all its derivatives and the ChEMBL26 compounds dataset. Conditions, including activity thresholds and input features, were explored to access the performance of eight machine learning methods of target prediction of NPs, including support vector machines (SVM), extreme gradient boosting, random forests, K-nearest neighbor, naive Bayes, feedforward neural networks (FNN), convolutional neural networks and recurrent neural networks. As a result, the NPs and all their derivatives datasets were selected to build the best NP-specific models. Furthermore, the consensus models, as well as the voting models, were additionally applied to improve the prediction performance. More evaluations were made on the external validation set and the results demonstrated that (1) the NP-specific model performed better on the target prediction of NPs than the traditional models training on the whole compounds of ChEMBL26. (2) The consensus model of FNN + SVM possessed the best overall performance, and the voting model can significantly improve recall and specificity.

Spatiotemporal Dynamics of Stress-Induced Network Reconfigurations Reflect Negative Affectivity

BACKGROUND

Maladaptive stress responses are important risk factors in the etiology of mood and anxiety disorders, but exact pathomechanisms remain to be understood. Mapping individual differences of acute stress-induced neurophysiological changes, especially on the level of neural activation and functional connectivity (FC), could provide important insights in how variation in the individual stress response is linked to disease risk.

METHODS

Using an established psychosocial stress task flanked by two resting states, we measured subjective, physiological, and brain responses to acute stress and recovery in 217 participants with and without mood and anxiety disorders. To estimate blockwise changes in stress-induced activation and FC, we used hierarchical mixed-effects models based on denoised time series within predefined stress-related regions. We predicted inter- and intraindividual differences in stress phases (anticipation vs. stress vs. recovery) and transdiagnostic dimensions of stress reactivity using elastic net and support vector machines.

RESULTS

We identified four subnetworks showing distinct changes in FC over time. FC but not activation trajectories predicted the stress phase (accuracy = 70%, p_perm < .001) and increases in heart rate (R² = 0.075, p_perm < .001). Critically, individual spatiotemporal trajectories of changes across networks also predicted negative affectivity (ΔR² = 0.075, p_perm = .030) but not the presence or absence of a mood and anxiety disorder.

CONCLUSIONS

Spatiotemporal dynamics of brain network reconfiguration induced by stress reflect individual differences in the psychopathology dimension of negative affectivity. These results support the idea that vulnerability for mood and anxiety disorders can be conceptualized best at the level of network dynamics, which may pave the way for improved prediction of individual risk.

Mind the gap: Performance metric evaluation in brain-age prediction

Estimating age based on neuroimaging-derived data has become a popular approach to developing markers for brain integrity and health. While a variety of machine-learning algorithms can provide accurate predictions of age based on brain characteristics, there is significant variation in model accuracy reported across studies. We predicted age in two population-based datasets, and assessed the effects of age range, sample size and age-bias correction on the model performance metrics Pearson's correlation coefficient (r), the coefficient of determination (R2 ), Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). The results showed that these metrics vary considerably depending on cohort age range; r and R2 values are lower when measured in samples with a narrower age range. RMSE and MAE are also lower in samples with a narrower age range due to smaller errors/brain age delta values when predictions are closer to the mean age of the group. Across subsets with different age ranges, performance metrics improve with increasing sample size. Performance metrics further vary depending on prediction variance as well as mean age difference between training and test sets, and age-bias corrected metrics indicate high accuracy-also for models showing poor initial performance. In conclusion, performance metrics used for evaluating age prediction models depend on cohort and study-specific data characteristics, and cannot be directly compared across different studies. Since age-bias corrected metrics generally indicate high accuracy, even for poorly performing models, inspection of uncorrected model results provides important information about underlying model attributes such as prediction variance.