Assessing and tuning brain decoders: Cross-validation, caveats, and guidelines

Multiscale brain age prediction reveals region-specific accelerated brain aging in Parkinson's disease

Brain biological age, which measures the aging process in the brain using neuroimaging data, has been used to assess advanced brain aging in neurodegenerative diseases, including Parkinson disease (PD). However, assuming that whole brain degeneration is uniform may not be sufficient for assessing the complex neurodegenerative processes in PD. In this study we constructed a multiscale brain age prediction models based on structural MRI of 1240 healthy participants. To assess the brain aging patterns using the brain age prediction model, 93 PD patients and 91 healthy controls matching for sex and age were included. We found increased global and regional brain age in PD patients. The advanced aging regions were predominantly noted in the frontal and temporal cortices, limbic system, basal ganglia, thalamus, and cerebellum. Furthermore, region-level rather than global brain age in PD patients was associated with disease severity. Our multiscale brain age prediction model could aid in the development of objective image-based biomarkers to detect advanced brain aging in neurodegenerative diseases.

Improving Predictability, Test-Retest Reliability and Generalisability of Brain-Wide Associations for Cognitive Abilities via Multimodal Stacking

Brain-wide association studies (BWASs) have attempted to relate cognitive abilities with brain phenotypes, but have been challenged by issues such as predictability, test-retest reliability, and cross-cohort generalisability. To tackle these challenges, we proposed "stacking" that combines brain magnetic resonance imaging of different modalities, from task-fMRI contrasts and functional connectivity during tasks and rest to structural measures, into one prediction model. We benchmarked the benefits of stacking, using the Human Connectome Projects: Young Adults and Aging and the Dunedin Multidisciplinary Health and Development Study. For predictability, stacked models led to out-of-sample r ∼.5-.6 when predicting cognitive abilities at the time of scanning and 36 years earlier. For test-retest reliability, stacked models reached an excellent level of reliability (ICC>.75), even when we stacked only task-fMRI contrasts together. For generalisability, a stacked model with non-task MRI built from one dataset significantly predicted cognitive abilities in other datasets. Altogether, stacking is a viable approach to undertake the three challenges of BWAS for cognitive abilities.

Migraine aura discrimination using machine learning: an fMRI study during ictal and interictal periods

Functional magnetic resonance imaging (fMRI) studies on migraine with aura are challenging due to the rarity of patients with triggered cases. This study optimized methodologies to explore differences in ictal and interictal spatiotemporal activation patterns based on visual stimuli using fMRI in two patients with unique aura triggers. Both patients underwent separate fMRI sessions during the ictal and interictal periods. The Gaussian Process Classifier (GPC) was used to differentiate these periods by employing a machine learning temporal embedding approach and spatiotemporal activation patterns based on visual stimuli. When restricted to visual and occipital regions, GPC had an improved performance, with accuracy rates for patients A and B of roughly 86-90% and 77-81%, respectively (p < 0.01). The algorithm effectively differentiated visual stimulation and rest periods and identified times when aura symptoms manifested, as evident from the varying predicted probabilities in the GPC models. These findings contribute to our understanding of the role of visual processing and brain activity patterns in migraine with aura and the significance of temporal embedding techniques in examining aura phenomena. This finding has implications for diagnostic tools and therapeutic techniques, especially for patients suffering from aura symptoms.

Brain activity of professional investors signals future stock performance

A major aspiration of investors is to better forecast stock performance. Interestingly, emerging "neuroforecasting" research suggests that brain activity associated with anticipatory reward relates to market behavior and population-wide preferences, including stock price dynamics. In this study, we extend these findings to professional investors processing comprehensive real-world information on stock investment options while making predictions of long-term stock performance. Using functional MRI, we sampled investors' neural responses to investment cases and assessed whether these responses relate to future performance on the stock market. We found that our sample of investors could not successfully predict future market performance of the investment cases, confirming that stated preferences do not predict the market. Stock metrics of the investment cases were not predictive of future stock performance either. However, as investors processed case information, nucleus accumbens (NAcc) activity was higher for investment cases that ended up overperforming in the market. These findings remained robust, even when controlling for stock metrics and investors' predictions made in the scanner. Cross-validated prediction analysis indicated that NAcc activity could significantly predict future stock performance out-of-sample above chance. Our findings resonate with recent neuroforecasting studies and suggest that brain activity of professional investors may help in forecasting future stock performance.

Enhancing Cognitive Performance Prediction through White Matter Hyperintensity Connectivity Assessment: A Multicenter Lesion Network Mapping Analysis of 3,485 Memory Clinic Patients

Introduction

White matter hyperintensities of presumed vascular origin (WMH) are associated with cognitive impairment and are a key imaging marker in evaluating cognitive health. However, WMH volume alone does not fully account for the extent of cognitive deficits and the mechanisms linking WMH to these deficits remain unclear. We propose that lesion network mapping (LNM), enables to infer if brain networks are connected to lesions, and could be a promising technique for enhancing our understanding of the role of WMH in cognitive disorders. Our study employed this approach to test the following hypotheses: (1) LNM-informed markers surpass WMH volumes in predicting cognitive performance, and (2) WMH contributing to cognitive impairment map to specific brain networks.

Methods & results

We analyzed cross-sectional data of 3,485 patients from 10 memory clinic cohorts within the Meta VCI Map Consortium, using harmonized test results in 4 cognitive domains and WMH segmentations. WMH segmentations were registered to a standard space and mapped onto existing normative structural and functional brain connectome data. We employed LNM to quantify WMH connectivity across 480 atlas-based gray and white matter regions of interest (ROI), resulting in ROI-level structural and functional LNM scores. The capacity of total and regional WMH volumes and LNM scores in predicting cognitive function was compared using ridge regression models in a nested cross-validation. LNM scores predicted performance in three cognitive domains (attention and executive function, information processing speed, and verbal memory) significantly better than WMH volumes. LNM scores did not improve prediction for language functions. ROI-level analysis revealed that higher LNM scores, representing greater disruptive effects of WMH on regional connectivity, in gray and white matter regions of the dorsal and ventral attention networks were associated with lower cognitive performance.

Conclusion

Measures of WMH-related brain network connectivity significantly improve the prediction of current cognitive performance in memory clinic patients compared to WMH volume as a traditional imaging marker of cerebrovascular disease. This highlights the crucial role of network effects, particularly in attentionrelated brain regions, improving our understanding of vascular contributions to cognitive impairment. Moving forward, refining WMH information with connectivity data could contribute to patient-tailored therapeutic interventions and facilitate the identification of subgroups at risk of cognitive disorders.

Brain decoding of spontaneous thought: Predictive modeling of self-relevance and valence using personal narratives

The contents and dynamics of spontaneous thought are important factors for personality traits and mental health. However, assessing spontaneous thoughts is challenging due to their unconstrained nature, and directing participants' attention to report their thoughts may fundamentally alter them. Here, we aimed to decode two key content dimensions of spontaneous thought-self-relevance and valence-directly from brain activity. To train functional MRI-based predictive models, we used individually generated personal stories as stimuli in a story-reading task to mimic narrative-like spontaneous thoughts (n = 49). We then tested these models on multiple test datasets (total n = 199). The default mode, ventral attention, and frontoparietal networks played key roles in the predictions, with the anterior insula and midcingulate cortex contributing to self-relevance prediction and the left temporoparietal junction and dorsomedial prefrontal cortex contributing to valence prediction. Overall, this study presents brain models of internal thoughts and emotions, highlighting the potential for the brain decoding of spontaneous thought.

Flexibility of brain dynamics is increased and predicts clinical impairment in relapsing-remitting but not in secondary progressive multiple sclerosis

Large-scale brain activity has long been investigated under the erroneous assumption of stationarity. Nowadays, we know that resting-state functional connectivity is characterized by aperiodic, scale-free bursts of activity (i.e. neuronal avalanches) that intermittently recruit different brain regions. These different patterns of activity represent a measure of brain flexibility, whose reduction has been found to predict clinical impairment in multiple neurodegenerative diseases such as Parkinson's disease, amyotrophic lateral sclerosis and Alzheimer's disease. Brain flexibility has been recently found increased in multiple sclerosis, but its relationship with clinical disability remains elusive. Also, potential differences in brain dynamics according to the multiple sclerosis clinical phenotypes remain unexplored so far. We performed a brain dynamics study quantifying brain flexibility utilizing the 'functional repertoire' (i.e. the number of configurations of active brain areas) through source reconstruction of magnetoencephalography signals in a cohort of 25 multiple sclerosis patients (10 relapsing-remitting multiple sclerosis and 15 secondary progressive multiple sclerosis) and 25 healthy controls. Multiple sclerosis patients showed a greater number of unique reconfigurations at fast time scales as compared with healthy controls. This difference was mainly driven by the relapsing-remitting multiple sclerosis phenotype, whereas no significant differences in brain dynamics were found between secondary progressive multiple sclerosis and healthy controls. Brain flexibility also showed a different predictive power on clinical disability according to the multiple sclerosis type. For the first time, we investigated brain dynamics in multiple sclerosis patients through high temporal resolution techniques, unveiling differences in brain flexibility according to the multiple sclerosis phenotype and its relationship with clinical disability.

Subject-independent meta-learning framework towards optimal training of EEG-based classifiers

Advances in deep learning have shown great promise towards the application of performing high-accuracy Electroencephalography (EEG) signal classification in a variety of tasks. However, many EEG-based datasets are often plagued by the issue of high inter-subject signal variability. Robust deep learning models are notoriously difficult to train under such scenarios, often leading to subpar or widely varying performance across subjects under the leave-one-subject-out paradigm. Recently, the model agnostic meta-learning framework was introduced as a way to increase the model's ability to generalize towards new tasks. While the original framework focused on task-based meta-learning, this research aims to show that the meta-learning methodology can be modified towards subject-based signal classification while maintaining the same task objectives and achieve state-of-the-art performance. Namely, we propose the novel implementation of a few/zero-shot subject-independent meta-learning framework towards multi-class inner speech and binary class motor imagery classification. Compared to current subject-adaptive methods which utilize large number of labels from the target, the proposed framework shows its effectiveness in training zero-calibration and few-shot models for subject-independent EEG classification. The proposed few/zero-shot subject-independent meta-learning mechanism performs well on both small and large datasets and achieves robust, generalized performance across subjects. The results obtained shows a significant improvement over the current state-of-the-art, with the binary class motor imagery achieving 88.70% and the accuracy of multi-class inner speech achieving an average of 31.15%. Codes will be made available to public upon publication.

Linguistic modulation of the neural encoding of phonemes

Speech comprehension entails the neural mapping of the acoustic speech signal onto learned linguistic units. This acousto-linguistic transformation is bi-directional, whereby higher-level linguistic processes (e.g. semantics) modulate the acoustic analysis of individual linguistic units. Here, we investigated the cortical topography and linguistic modulation of the most fundamental linguistic unit, the phoneme. We presented natural speech and "phoneme quilts" (pseudo-randomly shuffled phonemes) in either a familiar (English) or unfamiliar (Korean) language to native English speakers while recording functional magnetic resonance imaging. This allowed us to dissociate the contribution of acoustic vs. linguistic processes toward phoneme analysis. We show that (i) the acoustic analysis of phonemes is modulated by linguistic analysis and (ii) that for this modulation, both of acoustic and phonetic information need to be incorporated. These results suggest that the linguistic modulation of cortical sensitivity to phoneme classes minimizes prediction error during natural speech perception, thereby aiding speech comprehension in challenging listening situations.

Strength of spatial correlation between gray matter connectivity and patterns of proto-oncogene and neural network construction gene expression is associated with diffuse glioma survival

Introduction

Like other forms of neuropathology, gliomas appear to spread along neural pathways. Accordingly, our group and others have previously shown that brain network connectivity is highly predictive of glioma survival. In this study, we aimed to examine the molecular mechanisms of this relationship via imaging transcriptomics.

Methods

We retrospectively obtained presurgical, T1-weighted MRI datasets from 669 adult patients, newly diagnosed with diffuse glioma. We measured brain connectivity using gray matter networks and coregistered these data with a transcriptomic brain atlas to determine the spatial co-localization between brain connectivity and expression patterns for 14 proto-oncogenes and 3 neural network construction genes.

Results

We found that all 17 genes were significantly co-localized with brain connectivity (p < 0.03, corrected). The strength of co-localization was highly predictive of overall survival in a cross-validated Cox Proportional Hazards model (mean area under the curve, AUC = 0.68 +/- 0.01) and significantly (p < 0.001) more so for a random forest survival model (mean AUC = 0.97 +/- 0.06). Bayesian network analysis demonstrated direct and indirect causal relationships among gene-brain co-localizations and survival. Gene ontology analysis showed that metabolic processes were overexpressed when spatial co-localization between brain connectivity and gene transcription was highest (p < 0.001). Drug-gene interaction analysis identified 84 potential candidate therapies based on our findings.

Discussion

Our findings provide novel insights regarding how gene-brain connectivity interactions may affect glioma survival.

The genetic architecture of multimodal human brain age

The complex biological mechanisms underlying human brain aging remain incompletely understood. This study investigated the genetic architecture of three brain age gaps (BAG) derived from gray matter volume (GM-BAG), white matter microstructure (WM-BAG), and functional connectivity (FC-BAG). We identified sixteen genomic loci that reached genome-wide significance (P-value < 5×10-8). A gene-drug-disease network highlighted genes linked to GM-BAG for treating neurodegenerative and neuropsychiatric disorders and WM-BAG genes for cancer therapy. GM-BAG displayed the most pronounced heritability enrichment in genetic variants within conserved regions. Oligodendrocytes and astrocytes, but not neurons, exhibited notable heritability enrichment in WM and FC-BAG, respectively. Mendelian randomization identified potential causal effects of several chronic diseases on brain aging, such as type 2 diabetes on GM-BAG and AD on WM-BAG. Our results provide insights into the genetics of human brain aging, with clinical implications for potential lifestyle and therapeutic interventions. All results are publicly available at https://labs.loni.usc.edu/medicine .

Neurodevelopmental Subtypes of Functional Brain Organization in the ABCD Study Using a Rigorous Analytic Framework

The current study demonstrates that an individual's resting-state functional connectivity (RSFC) is a dependable biomarker for identifying differential patterns of cognitive and emotional functioning during late childhood. Using baseline RSFC data from the Adolescent Brain Cognitive Development (ABCD) study, which includes children aged 9-11, we identified four distinct RSFC subtypes We introduce an integrated methodological pipeline for testing the reliability and importance of these subtypes. In the Identification phase, Leiden Community Detection defined RSFC subtypes, with their reproducibility confirmed through a split-sample technique in the Validation stage. The Evaluation phase showed that distinct cognitive and mental health profiles are associated with each subtype, with the Predictive phase indicating that subtypes better predict various cognitive and mental health characteristics than individual RSFC connections. The Replication stage employed bootstrapping and down-sampling methods to substantiate the reproducibility of these subtypes further. This work allows future explorations of developmental trajectories of these RSFC subtypes.

Toward Generalizable Machine Learning Models in Speech, Language, and Hearing Sciences: Estimating Sample Size and Reducing Overfitting

PURPOSE

Many studies using machine learning (ML) in speech, language, and hearing sciences rely upon cross-validations with single data splitting. This study's first purpose is to provide quantitative evidence that would incentivize researchers to instead use the more robust data splitting method of nested k-fold cross-validation. The second purpose is to present methods and MATLAB code to perform power analysis for ML-based analysis during the design of a study.

METHOD

First, the significant impact of different cross-validations on ML outcomes was demonstrated using real-world clinical data. Then, Monte Carlo simulations were used to quantify the interactions among the employed cross-validation method, the discriminative power of features, the dimensionality of the feature space, the dimensionality of the model, and the sample size. Four different cross-validation methods (single holdout, 10-fold, train-validation-test, and nested 10-fold) were compared based on the statistical power and confidence of the resulting ML models. Distributions of the null and alternative hypotheses were used to determine the minimum required sample size for obtaining a statistically significant outcome (5% significance) with 80% power. Statistical confidence of the model was defined as the probability of correct features being selected for inclusion in the final model.

RESULTS

ML models generated based on the single holdout method had very low statistical power and confidence, leading to overestimation of classification accuracy. Conversely, the nested 10-fold cross-validation method resulted in the highest statistical confidence and power while also providing an unbiased estimate of accuracy. The required sample size using the single holdout method could be 50% higher than what would be needed if nested k-fold cross-validation were used. Statistical confidence in the model based on nested k-fold cross-validation was as much as four times higher than the confidence obtained with the single holdout-based model. A computational model, MATLAB code, and lookup tables are provided to assist researchers with estimating the minimum sample size needed during study design.

CONCLUSION

The adoption of nested k-fold cross-validation is critical for unbiased and robust ML studies in the speech, language, and hearing sciences.

SUPPLEMENTAL MATERIAL

https://doi.org/10.23641/asha.25237045.

On decoding of rapid motor imagery in a diverse population using a high-density NIRS device

Introduction

Functional near-infrared spectroscopy (fNIRS) aims to infer cognitive states such as the type of movement imagined by a study participant in a given trial using an optical method that can differentiate between oxygenation states of blood in the brain and thereby indirectly between neuronal activity levels. We present findings from an fNIRS study that aimed to test the applicability of a high-density (>3000 channels) NIRS device for use in short-duration (2 s) left/right hand motor imagery decoding in a diverse, but not explicitly balanced, subject population. A side aim was to assess relationships between data quality, self-reported demographic characteristics, and brain-computer interface (BCI) performance, with no subjects rejected from recruitment or analysis.

Methods

BCI performance was quantified using several published methods, including subject-specific and subject-independent approaches, along with a high-density fNIRS decoder previously validated in a separate study.

Results

We found that decoding of motor imagery on this population proved extremely challenging across all tested methods. Overall accuracy of the best-performing method (the high-density decoder) was 59.1 +/- 6.7% after excluding subjects where almost no optode-scalp contact was made over motor cortex and 54.7 +/- 7.6% when all recorded sessions were included. Deeper investigation revealed that signal quality, hemodynamic responses, and BCI performance were all strongly impacted by the hair phenotypical and demographic factors under investigation, with over half of variance in signal quality explained by demographic factors alone.

Discussion

Our results contribute to the literature reporting on challenges in using current-generation NIRS devices on subjects with long, dense, dark, and less pliable hair types along with the resulting potential for bias. Our findings confirm the need for increased focus on these populations, accurate reporting of data rejection choices across subject intake, curation, and final analysis in general, and signal a need for NIRS optode designs better optimized for the general population to facilitate more robust and inclusive research outcomes.

Neural dynamics of predictive timing and motor engagement in music listening

Why do humans spontaneously dance to music? To test the hypothesis that motor dynamics reflect predictive timing during music listening, we created melodies with varying degrees of rhythmic predictability (syncopation) and asked participants to rate their wanting-to-move (groove) experience. Degree of syncopation and groove ratings are quadratically correlated. Magnetoencephalography data showed that, while auditory regions track the rhythm of melodies, beat-related 2-hertz activity and neural dynamics at delta (1.4 hertz) and beta (20 to 30 hertz) rates in the dorsal auditory pathway code for the experience of groove. Critically, the left sensorimotor cortex coordinates these groove-related delta and beta activities. These findings align with the predictions of a neurodynamic model, suggesting that oscillatory motor engagement during music listening reflects predictive timing and is effected by interaction of neural dynamics along the dorsal auditory pathway.

Julearn: an easy-to-use library for leakage-free evaluation and inspection of ML models

The fast-paced development of machine learning (ML) and its increasing adoption in research challenge researchers without extensive training in ML. In neuroscience, ML can help understand brain-behavior relationships, diagnose diseases and develop biomarkers using data from sources like magnetic resonance imaging and electroencephalography. Primarily, ML builds models to make accurate predictions on unseen data. Researchers evaluate models' performance and generalizability using techniques such as cross-validation (CV). However, choosing a CV scheme and evaluating an ML pipeline is challenging and, if done improperly, can lead to overestimated results and incorrect interpretations. Here, we created julearn, an open-source Python library allowing researchers to design and evaluate complex ML pipelines without encountering common pitfalls. We present the rationale behind julearn's design, its core features, and showcase three examples of previously-published research projects. Julearn simplifies the access to ML providing an easy-to-use environment. With its design, unique features, simple interface, and practical documentation, it poses as a useful Python-based library for research projects.

Data leakage inflates prediction performance in connectome-based machine learning models

Predictive modeling is a central technique in neuroimaging to identify brain-behavior relationships and test their generalizability to unseen data. However, data leakage undermines the validity of predictive models by breaching the separation between training and test data. Leakage is always an incorrect practice but still pervasive in machine learning. Understanding its effects on neuroimaging predictive models can inform how leakage affects existing literature. Here, we investigate the effects of five forms of leakage-involving feature selection, covariate correction, and dependence between subjects-on functional and structural connectome-based machine learning models across four datasets and three phenotypes. Leakage via feature selection and repeated subjects drastically inflates prediction performance, whereas other forms of leakage have minor effects. Furthermore, small datasets exacerbate the effects of leakage. Overall, our results illustrate the variable effects of leakage and underscore the importance of avoiding data leakage to improve the validity and reproducibility of predictive modeling.

Brain Functional Connectivity and Anatomical Features as Predictors of Cognitive Behavioral Therapy Outcome for Anxiety in Youths

Background

Because pediatric anxiety disorders precede the onset of many other problems, successful prediction of response to the first-line treatment, cognitive-behavioral therapy (CBT), could have major impact. However, existing clinical models are weakly predictive. The current study evaluates whether structural and resting-state functional magnetic resonance imaging can predict post-CBT anxiety symptoms.

Methods

Two datasets were studied: (A) one consisted of n=54 subjects with an anxiety diagnosis, who received 12 weeks of CBT, and (B) one consisted of n=15 subjects treated for 8 weeks. Connectome Predictive Modeling (CPM) was used to predict treatment response, as assessed with the PARS; additionally we investigated models using anatomical features, instead of functional connectivity. The main analysis included network edges positively correlated with treatment outcome, and age, sex, and baseline anxiety severity as predictors. Results from alternative models and analyses also are presented. Model assessments utilized 1000 bootstraps, resulting in a 95% CI for R2, r and mean absolute error (MAE).

Outcomes

The main model showed a mean absolute error of approximately 3.5 (95%CI: [3.1-3.8]) points a R2 of 0.08 [-0.14 - 0.26] and r of 0.38 [0.24 - 0.511]. When testing this model in the left-out sample (B) the results were similar, with a MAE of 3.4 [2.8 - 4.7], R2-0.65 [-2.29 - 0.16] and r of 0.4 [0.24 - 0.54]. The anatomical metrics showed a similar pattern, where models rendered overall low R2.

Interpretation

The analysis showed that models based on earlier promising results failed to predict clinical outcomes. Despite the small sample size, the current study does not support extensive use of CPM to predict outcome in pediatric anxiety.

The effect of dopaminergic treatment on whole body kinematics explored through network theory

Three-dimensional motion analysis represents a quantitative approach to assess spatio-temporal and kinematic changes in health and disease. However, these parameters provide only segmental information, discarding minor changes of complex whole body kinematics characterizing physiological and/or pathological conditions. We aimed to assess how levodopa intake affects the whole body, analyzing the kinematic interactions during gait in Parkinson's disease (PD) through network theory which assess the relationships between elements of a system. To this end, we analysed gait data of 23 people with PD applying network theory to the acceleration kinematic data of 21 markers placed on participants' body landmarks. We obtained a matrix of kinematic interactions (i.e., the kinectome) for each participant, before and after the levodopa intake, we performed a topological analysis to evaluate the large-scale interactions among body elements, and a multilinear regression analysis to verify whether the kinectome's topology could predict the clinical variations induced by levodopa. We found that, following levodopa intake, patients with PD showed less trunk and head synchronization (p-head = 0.048; p-7th cervical vertebrae = 0.032; p-10th thoracic vertebrae = 0.006) and an improved upper-lower limbs synchronization (elbows right, p = 0.002; left, p = 0.005), (wrists right, p = 0.003; left, p = 0.002; knees right, p = 0.003; left, p = 0.039) proportional to the UPDRS-III scores. These results may be attributable to the reduction of rigidity, following pharmacological treatment.

Investigation of different ML approaches in classification of emotions induced by acute stress

Background

Machine learning is becoming a common tool in monitoring emotion. However, methodological studies of the processing pipeline are scarce, especially ones using subjective appraisals as ground truth.

New method

A novel protocol was used to induce cognitive load and physical discomfort, and emotional dimensions (arousal, valence, and dominance) were reported after each task. The performance of five common ML models with a versatile set of features (physiological features, task performance data, and personality trait) was compared in binary classification of subjectively assessed emotions.

Results

The psychophysiological responses proved the protocol was successful in changing the mental state from baseline, also the cognitive and physical tasks were different. The optimization and performance of ML models used for emotion detection were evaluated. Additionally, methods to account for imbalanced classes were applied and shown to improve the classification performance.

Comparison with existing methods

Classification of human emotional states often assumes the states are determined by the stimuli. However, individual appraisals vary. None of the past studies have classified subjective emotional dimensions with a set of features including biosignals, personality and behavior.

Conclusion

Our data represent a typical setup in affective computing utilizing psychophysiological monitoring: N is low compared to number of features, inter-individual variability is high, and class imbalance cannot be avoided. Our observations are a) if possible, include features representing physiology, behavior and personality, b) use simple models and limited number of features to improve interpretability, c) address the possible imbalance, d) if the data size allows, use nested cross-validation.

Practical challenges for precision medicine

The prediction of individual treatment responses with machine learning faces hurdles.

Solving large-scale MEG/EEG source localisation and functional connectivity problems simultaneously using state-space models

State-space models are widely employed across various research disciplines to study unobserved dynamics. Conventional estimation techniques, such as Kalman filtering and expectation maximisation, offer valuable insights but incur high computational costs in large-scale analyses. Sparse inverse covariance estimators can mitigate these costs, but at the expense of a trade-off between enforced sparsity and increased estimation bias, necessitating careful assessment in low signal-to-noise ratio (SNR) situations. To address these challenges, we propose a three-fold solution: (1) Introducing multiple penalised state-space (MPSS) models that leverage data-driven regularisation; (2) Developing novel algorithms derived from backpropagation, gradient descent, and alternating least squares to solve MPSS models; (3) Presenting a K-fold cross-validation extension for evaluating regularisation parameters. We validate this MPSS regularisation framework through lower and more complex simulations under varying SNR conditions, including a large-scale synthetic magneto- and electro-encephalography (MEG/EEG) data analysis. In addition, we apply MPSS models to concurrently solve brain source localisation and functional connectivity problems for real event-related MEG/EEG data, encompassing thousands of sources on the cortical surface. The proposed methodology overcomes the limitations of existing approaches, such as constraints to small-scale and region-of-interest analyses. Thus, it may enable a more accurate and detailed exploration of cognitive brain functions.

The effects of data leakage on connectome-based machine learning models

Predictive modeling has now become a central technique in neuroimaging to identify complex brain-behavior relationships and test their generalizability to unseen data. However, data leakage, which unintentionally breaches the separation between data used to train and test the model, undermines the validity of predictive models. Previous literature suggests that leakage is generally pervasive in machine learning, but few studies have empirically evaluated the effects of leakage in neuroimaging data. Although leakage is always an incorrect practice, understanding the effects of leakage on neuroimaging predictive models provides insight into the extent to which leakage may affect the literature. Here, we investigated the effects of leakage on machine learning models in two common neuroimaging modalities, functional and structural connectomes. Using over 400 different pipelines spanning four large datasets and three phenotypes, we evaluated five forms of leakage fitting into three broad categories: feature selection, covariate correction, and lack of independence between subjects. As expected, leakage via feature selection and repeated subjects drastically inflated prediction performance. Notably, other forms of leakage had only minor effects (e.g., leaky site correction) or even decreased prediction performance (e.g., leaky covariate regression). In some cases, leakage affected not only prediction performance, but also model coefficients, and thus neurobiological interpretations. Finally, we found that predictive models using small datasets were more sensitive to leakage. Overall, our results illustrate the variable effects of leakage on prediction pipelines and underscore the importance of avoiding data leakage to improve the validity and reproducibility of predictive modeling.

Clustering of antipsychotic-naïve patients with schizophrenia based on functional connectivity from resting-state electroencephalography

Schizophrenia is associated with aberrations in the Default Mode Network (DMN), but the clinical implications remain unclear. We applied data-driven, unsupervised machine learning based on resting-state electroencephalography (rsEEG) functional connectivity within the DMN to cluster antipsychotic-naïve patients with first-episode schizophrenia. The identified clusters were investigated with respect to psychopathological profile and cognitive deficits. Thirty-seven antipsychotic-naïve, first-episode patients with schizophrenia (mean age 24.4 (5.4); 59.5% males) and 97 matched healthy controls (mean age 24.0 (5.1); 52.6% males) underwent assessments of rsEEG, psychopathology, and cognition. Source-localized, frequency-dependent functional connectivity was estimated using Phase Lag Index (PLI). The DMN-PLI was factorized for each frequency band using principal component analysis. Clusters of patients were identified using a Gaussian mixture model and neurocognitive and psychopathological profiles of identified clusters were explored. We identified two clusters of patients based on the theta band (4-8 Hz), and two clusters based on the beta band (12-30 Hz). Baseline psychopathology could predict theta clusters with an accuracy of 69.4% (p = 0.003), primarily driven by negative symptoms. Five a priori selected cognitive functions conjointly predicted the beta clusters with an accuracy of 63.6% (p = 0.034). The two beta clusters displayed higher and lower DMN connectivity, respectively, compared to healthy controls. In conclusion, the functional connectivity within the DMN provides a novel, data-driven means to stratify patients into clinically relevant clusters. The results support the notion of biological subgroups in schizophrenia and endorse the application of data-driven methods to recognize pathophysiological patterns at earliest stage of this syndrome.

Connectome-based prediction of craving in gambling disorder and cocaine use disorder

INTRODUCTION

Craving, involving intense and urgent desires to engage in specific behaviours, is a feature of addictions. Multiple studies implicate regions of salience/limbic networks and basal ganglia, fronto-parietal, medial frontal regions in craving in addictions. However, prior studies have not identified common neural networks that reliably predict craving across substance and behavioural addictions.

METHODS

Functional magnetic resonance imaging during an audiovisual cue-reactivity task and connectome-based predictive modelling (CPM), a data-driven method for generating brain-behavioural models, were used to study individuals with cocaine-use disorder and gambling disorder. Functions of nodes and networks relevant to craving were identified and interpreted based on meta-analytic data.

RESULTS

Craving was predicted by neural connectivity across disorders. The highest degree nodes were mostly located in the prefrontal cortex. Overall, the prediction model included complex networks including motor/sensory, fronto-parietal, and default-mode networks. The decoding revealed high functional associations with components of memory, valence ratings, physiological responses, and finger movement/motor imagery.

CONCLUSIONS

Craving could be predicted across substance and behavioural addictions. The model may reflect general neural mechanisms of craving despite specificities of individual disorders. Prefrontal regions associated with working memory and autobiographical memory seem important in predicting craving. For further validation, the model should be tested in diverse samples and contexts.

3DCNN predicting brain age using diffusion tensor imaging

Neuroimaging-based brain age prediction using deep learning is gaining popularity. However, few studies have attempted to leverage diffusion tensor imaging (DTI) to predict brain age. In this study, we proposed a 3D convolutional neural network model (3DCNN) and trained it on fractional anisotropy (FA) data from six publicly available datasets (n = 2406, age = 17-60) to estimate brain age. Implementing a two-loop nested cross-validation scheme with a tenfold cross-validation procedure, we achieved a robust prediction performance of a mean absolute error (MAE) of 2.785 and a correlation coefficient of 0.932. We also employed Grad-Cam++ to visualize the salient features of the proposed model. We identified a few highly salient fiber tracts, including the genu of corpus callosum and the left cerebellar peduncle, among others that play a pivotal role in our model. In sum, our model reliably predicted brain age and provided novel insight into age-related changes in brains' axonal structure.

Social support and overeating in young women: The role of altering functional network connectivity patterns and negative emotions

Research suggests that social support has a protective effect on emotional health and emotionally induced overeating. Women are especially more sensitive to benefits from social support when facing eating problems. Although it has been demonstrated that social support can affect the neural processes of emotion regulation and reward perception, it is unclear how social support alters synergistic patterns in large-scale brain networks associated with negative emotions and overeating. We used a large sample of young women aged 17-22 years (N = 360) to examine how social support influences the synchrony of five intrinsic networks (executive control network [ECN], default mode network, salience network [SN], basal ganglia network, and precuneus network [PN]) and how these networks influence negative affect and overeating. Additionally, we explored these analyses in another sample of males (N = 136). After statistically controlling for differences in age and head movement, we observed significant associations of higher levels of social support with increased intra- and inter-network functional synchrony, particularly for ECN-centered network connectivity. Subsequent chain-mediated analyses showed that social support predicted overeating through the ECN-SN and ECN-PN network connectivity and negative emotions. However, these results were not found in men. These findings suggest that social support influences the synergistic patterns within and between intrinsic networks related to inhibitory control, emotion salience, self-referential thinking, and reward sensitivity. Furthermore, they reveal that social support and its neural markers may play a key role in young women's emotional health and eating behavior.

Topological changes of fast large-scale brain dynamics in mild cognitive impairment predict early memory impairment: a resting-state, source reconstructed, magnetoencephalography study

Functional connectivity has been used as a framework to investigate widespread brain interactions underlying cognitive deficits in mild cognitive impairment (MCI). However, many functional connectivity metrics focus on the average of the periodic activities, disregarding the aperiodic bursts of activity (i.e., the neuronal avalanches) characterizing the large-scale dynamic activities of the brain. Here, we apply the recently described avalanche transition matrix framework to source-reconstructed magnetoencephalography signals in a cohort of 32 MCI patients and 32 healthy controls to describe the spatio-temporal features of neuronal avalanches and explore their topological properties. Our results showed that MCI patients showed a more centralized network (as assessed by higher values of the degree divergence and leaf fraction) as compared to healthy controls. Furthermore, we found that the degree divergence (in the theta band) was predictive of hippocampal memory impairment. These findings highlight the role of the changes of aperiodic bursts in clinical conditions and may contribute to a more thorough phenotypical assessment of patients.

A neuromarker for deficit syndrome in schizophrenia from a combination of structural and functional magnetic resonance imaging

AIM

Deficit schizophrenia (DS), defined by primary and enduring negative symptoms, has been proposed as a promising homogeneous subtype of schizophrenia. It has been demonstrated that unimodal neuroimaging characteristics of DS were different from non-deficit schizophrenia (NDS), however, whether multimodal-based neuroimaging features could identify deficit syndrome remains to be determined.

METHODS

Functional and structural multimodal magnetic resonance imaging of DS, NDS and healthy controls were scanned. Voxel-based features of gray matter volume, fractional amplitude of low-frequency fluctuations, and regional homogeneity were extracted. The support vector machine classification models were constructed using these features separately and jointly. The most discriminative features were defined as the first 10% of features with the greatest weights. Moreover, relevance vector regression was applied to explore the predictive values of these top-weighted features in predicting negative symptoms.

RESULTS

The multimodal classifier achieved a higher accuracy (75.48%) compared with the single modal model in distinguishing DS from NDS. The most predictive brain regions were mainly located in the default mode and visual networks, exhibiting differences between functional and structural features. Further, the identified discriminative features significantly predicted scores of diminished expressivity factor in DS but not NDS.

CONCLUSIONS

The present study demonstrated that local properties of brain regions extracted from multimodal imaging data could distinguish DS from NDS with a machine learning-based approach and confirmed the relationship between distinctive features and the negative symptoms subdomain. These findings may improve the identification of potential neuroimaging signatures and improve the clinical assessment of the deficit syndrome.

Strength of spatial correlation between structural brain network connectivity and brain-wide patterns of proto-oncogene and neural network construction gene expression is associated with diffuse glioma survival

Like other forms of neuropathology, gliomas appear to spread along neural pathways. Accordingly, our group and others have previously shown that brain network connectivity is highly predictive of glioma survival. In this study, we aimed to examine the molecular mechanisms of this relationship via imaging transcriptomics. We retrospectively obtained presurgical, T1-weighted MRI datasets from 669 adult patients, newly diagnosed with diffuse glioma. We measured brain connectivity using gray matter networks and coregistered these data with a transcriptomic brain atlas to determine the spatial co-localization between brain connectivity and expression patterns for 14 proto-oncogenes and 3 neural network construction genes. We found that all 17 genes were significantly co-localized with brain connectivity (p < 0.03, corrected). The strength of co-localization was highly predictive of overall survival in a cross-validated Cox Proportional Hazards model (mean area under the curve, AUC = 0.68 +/- 0.01) and significantly (p < 0.001) more so for a random forest survival model (mean AUC = 0.97 +/- 0.06). Bayesian network analysis demonstrated direct and indirect causal relationships among gene-brain co-localizations and survival. Gene ontology analysis showed that metabolic processes were overexpressed when spatial co-localization between brain connectivity and gene transcription was highest (p < 0.001). Drug-gene interaction analysis identified 84 potential candidate therapies based on our findings. Our findings provide novel insights regarding how gene-brain connectivity interactions may affect glioma survival.

Neurocomputational mechanisms of biased impression formation in lonely individuals

Social impressions are fundamental in our daily interactions with other people but forming accurate impressions of our social partners can be biased to different extents. Loneliness has previously been suggested to induce biases that hinder the formation of accurate impressions of others for successful social bonding. Here, we demonstrated that despite counterfactual evidence, negative first impressions bias information weighting, leading to less favorable trustworthiness beliefs. Lonely individuals did not only have more negative expectations of others' social behavior, but they also manifested a stronger weighting bias. Reduced orbitofrontal cortex (OFC) activity was associated with a stronger weighting bias in lonelier individuals and mediated the relationship between loneliness and this weighting bias. Importantly, stronger coupling between OFC and temporoparietal junction compensated for such effects, promoting more positive trustworthiness beliefs especially in lonelier individuals. These findings bear potential for future basic and clinical investigations on social cognition and the development of clinical symptoms linked to loneliness.

Shared and distinct structural brain networks related to childhood maltreatment and social support: connectome-based predictive modeling

Childhood maltreatment (CM) has been associated with changes in structural brain connectivity even in the absence of mental illness. Social support, an important protective factor in the presence of childhood maltreatment, has been positively linked to white matter integrity. However, the shared effects of current social support and CM and their association with structural connectivity remain to be investigated. They might shed new light on the neurobiological basis of the protective mechanism of social support. Using connectome-based predictive modeling (CPM), we analyzed structural connectomes of N = 904 healthy adults derived from diffusion-weighted imaging. CPM predicts phenotypes from structural connectivity through a cross-validation scheme. Distinct and shared networks of white matter tracts predicting childhood trauma questionnaire scores and the social support questionnaire were identified. Additional analyses were applied to assess the stability of the results. CM and social support were predicted significantly from structural connectome data (all rs ≥ 0.119, all ps ≤ 0.016). Edges predicting CM and social support were inversely correlated, i.e., positively correlated with CM and negatively with social support, and vice versa, with a focus on frontal and temporal regions including the insula and superior temporal lobe. CPM reveals the predictive value of the structural connectome for CM and current social support. Both constructs are inversely associated with connectivity strength in several brain tracts. While this underlines the interconnectedness of these experiences, it suggests social support acts as a protective factor following adverse childhood experiences, compensating for brain network alterations. Future longitudinal studies should focus on putative moderating mechanisms buffering these adverse experiences.

Friend or foe: classifying collaborative interactions using fNIRS

To succeed, effective teams depend on both cooperative and competitive interactions between individual teammates. Depending on the context, cooperation and competition can amplify or neutralize a team's problem solving ability. Therefore, to assess successful collaborative problem solving, it is first crucial to distinguish competitive from cooperative interactions. We investigate the feasibility of using lightweight brain sensors to distinguish cooperative from competitive interactions in pairs of participants (N=84) playing a decision-making game involving uncertain outcomes. We measured brain activity using functional near-infrared spectroscopy (fNIRS) from social, motor, and executive areas during game play alone and in competition or cooperation with another participant. To distinguish competitive, cooperative, and alone conditions, we then trained support vector classifiers using combinations of features extracted from fNIRS data. We find that features from social areas of the brain outperform other features for discriminating competitive, cooperative, and alone conditions in cross-validation. Comparing the competitive and alone conditions, social features yield a 5% improvement over motor and executive features. Social features show promise as means of distinguishing competitive and cooperative environments in problem solving settings. Using fNIRS data provides a real-time measure of subjective experience in an ecologically valid environment. These results have the potential to inform intelligent team monitoring to provide better real-time feedback and improve team outcomes in naturalistic settings.

Association of loneliness and grey matter volume in the dorsolateral prefrontal cortex: the mediating role of interpersonal self-support traits

As a social and public health concern, loneliness is associated with an abundance of negative life outcomes such as depressive symptomatology, mortality, and sleep disturbance. Nevertheless, the neural basis underlying loneliness remains elusive; in addition, previous neuroimaging studies about loneliness mainly focused on the elderly and were limited by small sample sizes. Here, utilizing the voxel-based morphometry (VBM) approach via structural magnetic resonance imaging, we investigated the association between brain GMV and loneliness in 462 young adults (67.7% females, age = 18.59 ± 1.14 years). Results from whole-brain VBM analyses revealed that individuals with higher loneliness tended to have greater GMV in the right dorsolateral prefrontal cortex (DLPFC), which was thought to be associated with emotional regulation deficits and executive dysfunction. Importantly, the GMV-based predictive models (a machine-learning method) demonstrated that the correlation between loneliness and GMV in the DLPFC was robust. Further, interpersonal self-support traits (ISS), a Chinese indigenous personality construct and pivotal personality factor for resisting negative life outcomes, mediated the relationship between the GMV in the right DLPFC and loneliness. Taken together, the present study reveals that the GMV in right DLPFC acts as an underlying neurostructural correlate of loneliness in the healthy brain, and further provides a brain-personality-symptom pathway for protection against loneliness in which GMV of DLPFC affects loneliness through ISS traits. Future intervention procedures aiming to decrease loneliness and enhance mental health levels among young adults should be developed through improving interpersonal relationships such as social skills training.

Advanced brain age in community-dwelling population with combined physical and cognitive impairments

We investigated whether advanced brain biological age is associated with accelerated age-related physical and/or cognitive functional decline: mobility impairment no disability (MIND), cognitive impairment no dementia (CIND), and physio-cognitive decline syndrome (PCDS). We constructed a brain age prediction model using gray matter features from the magnetic resonance imaging of 1482 healthy individuals (aged 18-92 years). Predicted and chronological age differences were obtained (brain age gap [BAG]) and analyzed in another 1193 community-dwelling population aged ≥50 years. Among the 1193 participants, there were 501, 346, 148, and 198 in the robust, CIND, MIND, and PCDS groups, respectively. Participants with PCDS had significantly larger BAG (BAG = 2.99 ± 8.97) than the robust (BAG = -0.49 ± 9.27, p = 0.002; η2 = 0.014), CIND (BAG = 0.47 ± 9.16, p = 0.02; η2 = 0.01), and MIND (BAG = 0.36 ± 9.69, p = 0.036; η2 = 0.013) groups. Advanced brain aging is involved in the pathophysiology of the co-occurrence of physical and cognitive decline in the older people. The PCDS may be a clinical phenotype reflective of accelerated biological age in community-dwelling older individuals.

Modulation of resting-state functional connectivity in default mode network is associated with the long-term treatment outcome in major depressive disorder

BACKGROUND

Treatment non-response and recurrence are the main sources of disease burden in major depressive disorder (MDD). However, little is known about its neurobiological mechanism concerning the brain network changes accompanying pharmacotherapy. The present study investigated the changes in the intrinsic brain networks during 6-month antidepressant treatment phase associated with the treatment response and recurrence in MDD.

METHODS

Resting-state functional magnetic resonance imaging was acquired from untreated patients with MDD and healthy controls at baseline. The patients' depressive symptoms were monitored by using the Hamilton Rating Scale for Depression (HAMD). After 6 months of antidepressant treatment, patients were re-scanned and followed up every 6 months over 2 years. Traditional statistical analysis as well as machine learning approaches were conducted to investigate the longitudinal changes in macro-scale resting-state functional network connectivity (rsFNC) strength and micro-scale resting-state functional connectivity (rsFC) associated with long-term treatment outcome in MDD.

RESULTS

Repeated measures of the general linear model demonstrated a significant difference in the default mode network (DMN) rsFNC change before and after the 6-month antidepressant treatment between remitters and non-remitters. The difference in the rsFNC change over the 6-month antidepressant treatment between recurring and stable MDD was also specific to DMN. Machine learning analysis results revealed that only the DMN rsFC change successfully distinguished non-remitters from the remitters at 6 months and recurring from stable MDD during the 2-year follow-up.

CONCLUSION

Our findings demonstrated that the intrinsic DMN connectivity could be a unique and important target for treatment and recurrence prevention in MDD.

The Genetic Architecture of Multimodal Human Brain Age

The complex biological mechanisms underlying human brain aging remain incompletely understood, involving multiple body organs and chronic diseases. In this study, we used multimodal magnetic resonance imaging and artificial intelligence to examine the genetic architecture of the brain age gap (BAG) derived from gray matter volume (GM-BAG, N=31,557 European ancestry), white matter microstructure (WM-BAG, N=31,674), and functional connectivity (FC-BAG, N=32,017). We identified sixteen genomic loci that reached genome-wide significance (P-value<5×10-8). A gene-drug-disease network highlighted genes linked to GM-BAG for treating neurodegenerative and neuropsychiatric disorders and WM-BAG genes for cancer therapy. GM-BAG showed the highest heritability enrichment for genetic variants in conserved regions, whereas WM-BAG exhibited the highest heritability enrichment in the 5' untranslated regions; oligodendrocytes and astrocytes, but not neurons, showed significant heritability enrichment in WM and FC-BAG, respectively. Mendelian randomization identified potential causal effects of several exposure variables on brain aging, such as type 2 diabetes on GM-BAG (odds ratio=1.05 [1.01, 1.09], P-value=1.96×10-2) and AD on WM-BAG (odds ratio=1.04 [1.02, 1.05], P-value=7.18×10-5). Overall, our results provide valuable insights into the genetics of human brain aging, with clinical implications for potential lifestyle and therapeutic interventions. All results are publicly available at the MEDICINE knowledge portal: https://labs.loni.usc.edu/medicine.

Brain development mediates the relationship between self-reported poor parental monitoring and adolescent anxiety

Adolescence is the peak period for the onset of generalized anxiety disorder (GAD). Brain networks of cognitive and affective control in adolescents are not well developed when their exposure to external stimuli suddenly increases.Reasonable parental monitoring is especially important during this period.To examine the role of parental monitoring in the development of functional brain networks of GAD, we conducted a cross-validation-based predictive study based on the functional brain networks of 192 participants. We found that a set of functional brain networks, especially the default mode network and its connectivity with the frontoparietal network, could predict the ages of adolescents, which was replicated in three independent samples.Importantly, the difference between predicted age and chronological age significantly mediated the relationship between parental monitoring and anxiety levels. These findings suggest that inadequate parental monitoring plays a crucial role in the delayed development of specific brain networks associated with GAD in adolescents. Our work highlights the important role of parental monitoring in adolescent development.

The diagnostic performance of machine learning based on resting-state functional magnetic resonance imaging data for major depressive disorders: a systematic review and meta-analysis

Objective

Machine learning (ML) has been widely used to detect and evaluate major depressive disorder (MDD) using neuroimaging data, i.e., resting-state functional magnetic resonance imaging (rs-fMRI). However, the diagnostic efficiency is unknown. The aim of the study is to conduct an updated meta-analysis to evaluate the diagnostic performance of ML based on rs-fMRI data for MDD.

Methods

English databases were searched for relevant studies. The Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2) was used to assess the methodological quality of the included studies. A random-effects meta-analytic model was implemented to investigate the diagnostic efficiency, including sensitivity, specificity, diagnostic odds ratio (DOR), and area under the curve (AUC). Regression meta-analysis and subgroup analysis were performed to investigate the cause of heterogeneity.

Results

Thirty-one studies were included in this meta-analysis. The pooled sensitivity, specificity, DOR, and AUC with 95% confidence intervals were 0.80 (0.75, 0.83), 0.83 (0.74, 0.82), 14.00 (9, 22.00), and 0.86 (0.83, 0.89), respectively. Substantial heterogeneity was observed among the studies included. The meta-regression showed that the leave-one-out cross-validation (loocv) (sensitivity: p < 0.01, specificity: p < 0.001), graph theory (sensitivity: p < 0.05, specificity: p < 0.01), n > 100 (sensitivity: p < 0.001, specificity: p < 0.001), simens equipment (sensitivity: p < 0.01, specificity: p < 0.001), 3.0T field strength (Sensitivity: p < 0.001, specificity: p = 0.04), and Beck Depression Inventory (BDI) (sensitivity: p = 0.04, specificity: p = 0.06) might be the sources of heterogeneity. Furthermore, the subgroup analysis showed that the sample size (n > 100: sensitivity: 0.71, specificity: 0.72, n < 100: sensitivity: 0.81, specificity: 0.79), the different levels of disease evaluated by the Hamilton Depression Rating Scale (HDRS/HAMD) (mild vs. moderate vs. severe: sensitivity: 0.52 vs. 0.86 vs. 0.89, specificity: 0.62 vs. 0.78 vs. 0.82, respectively), the depression scales in patients with comparable levels of severity. (BDI vs. HDRS/HAMD: sensitivity: 0.86 vs. 0.87, specificity: 0.78 vs. 0.80, respectively), and the features (graph vs. functional connectivity: sensitivity: 0.84 vs. 0.86, specificity: 0.76 vs. 0.78, respectively) selected might be the causes of heterogeneity.

Conclusion

ML showed high accuracy for the automatic diagnosis of MDD. Future studies are warranted to promote the potential use of these classification algorithms in clinical settings.

Speech comprehension across time, space, frequency, and age: MEG-MVPA classification of intertrial phase coherence

Language is a key part of human cognition, essential for our well-being at all stages of our lives. Whereas many neurocognitive abilities decline with age, for language the picture is much less clear, and how exactly speech comprehension changes with ageing is still unknown. To investigate this, we employed magnetoencephalography (MEG) and recorded neuromagnetic brain responses to auditory linguistic stimuli in healthy participants of younger and older age using a passive task-free paradigm and a range of different linguistic stimulus contrasts, which enabled us to assess neural processing of spoken language at multiple levels (lexical, semantic, morphosyntactic). Using machine learning-based classification algorithms to scrutinise intertrial phase coherence of MEG responses in cortical source space, we found that patterns of oscillatory neural activity diverged between younger and older participants across several frequency bands (alpha, beta, gamma) for all tested linguistic information types. The results suggest multiple age-related changes in the brain's neurolinguistic circuits, which may be due to both healthy ageing in general and compensatory processes in particular.

Conceptual representations in the default, control and attention networks are task-dependent and cross-modal

Conceptual knowledge is central to human cognition. Neuroimaging studies suggest that conceptual processing involves modality-specific and multimodal brain regions in a task-dependent fashion. However, it remains unclear (1) to what extent conceptual feature representations are also modulated by the task, (2) whether conceptual representations in multimodal regions are indeed cross-modal, and (3) how the conceptual system relates to the large-scale functional brain networks. To address these issues, we conducted multivariate pattern analyses on fMRI data. 40 participants performed three tasks-lexical decision, sound judgment, and action judgment-on written words. We found that (1) conceptual feature representations are strongly modulated by the task, (2) conceptual representations in several multimodal regions are cross-modal, and (3) conceptual feature retrieval involves the default, frontoparietal control, and dorsal attention networks. Conceptual representations in these large-scale networks are task-dependent and cross-modal. Our findings support theories that assume conceptual processing to rely on a flexible, multi-level architecture.

Prior Knowledge Uses Prestimulus Alpha Band Oscillations and Persistent Poststimulus Neural Templates for Conscious Perception

Prior knowledge has a profound impact on the way we perceive the world. However, it remains unclear how the prior knowledge is maintained in our brains and thereby influences the subsequent conscious perception. The Dalmatian dog illusion is a perfect tool to study prior knowledge, where the picture is initially perceived as noise. Once the prior knowledge was introduced, a Dalmatian dog could be consciously seen, and the picture immediately became meaningful. Using pictures with hidden objects as standard stimuli and similar pictures without hidden objects as deviant stimuli, we investigated the neural representation of prior knowledge and its impact on conscious perception in an oddball paradigm using electroencephalogram (EEG) in both male and female human subjects. We found that the neural patterns between the prestimulus alpha band oscillations and poststimulus EEG activity were significantly more similar for the standard stimuli than for the deviant stimuli after prior knowledge was provided. Furthermore, decoding analysis revealed that persistent neural templates were evoked after the introduction of prior knowledge, similar to that evoked in the early stages of visual processing. In conclusion, the current study suggests that prior knowledge uses alpha band oscillations in a multivariate manner in the prestimulus period and induces specific persistent neural templates in the poststimulus period, enabling the conscious perception of the hidden objects.SIGNIFICANCE STATEMENT The visual world we live in is not always optimal. In dark or noisy environments, prior knowledge can help us interpret imperfect sensory signals and enable us to consciously perceive hidden objects. However, we still know very little about how prior knowledge works at the neural level. Using the Dalmatian dog illusion and multivariate methods, we found that prior knowledge uses prestimulus alpha band oscillations to carry information about the hidden object and exerts a persistent influence in the poststimulus period by inducing specific neural templates. Our findings provide a window into the neural underpinnings of prior knowledge and offer new insights into the role of alpha band oscillations and neural templates associated with conscious perception.

Connectome-based fingerprint of motor impairment is stable along the course of Parkinson's disease

Functional alterations in brain connectivity have previously been described in Parkinson's disease, but it is not clear whether individual differences in connectivity profiles might be also linked to severity of motor-symptom manifestation. Here we investigated the relevance of individual functional connectivity patterns measured with resting-state fMRI with respect to motor-symptom severity in Parkinson's disease, through a whole-brain, data-driven approach (connectome-based predictive modeling). Neuroimaging and clinical data of Parkinson's disease patients from the Parkinson's Progression Markers Initiative were derived at baseline (session 1, n = 81) and at follow-up (session 2, n = 53). Connectome-based predictive modeling protocol was implemented to predict levels of motor impairment from individual connectivity profiles. The resulting predictive model comprised a network mainly involving functional connections between regions located in the cerebellum, and in the motor and frontoparietal networks. The predictive power of the model was stable along disease progression, as the connectivity within the same network could predict levels of motor impairment, even at a later stage of the disease. Finally, connectivity profiles within this network could be identified at the individual level, suggesting the presence of individual fingerprints within resting-state fMRI connectivity associated with motor manifestations in Parkinson's disease.

Different patterns of intrinsic functional connectivity at the default mode and attentional networks predict crystalized and fluid abilities in childhood

Crystallized abilities are skills used to solve problems based on experience, while fluid abilities are linked to reasoning without evoke prior knowledge. To what extent crystallized and fluid abilities involve dissociated or overlapping neural systems is debatable. Due to often deployed small sample sizes or different study settings in prior work, the neural basis of crystallized and fluid abilities in childhood remains largely unknown. Here we analyzed within and between network connectivity patterns from resting-state functional MRI of 2707 children between 9 and 10 years from the ABCD study. We hypothesized that differences in functional connectivity at the default mode network (DMN), ventral, and dorsal attentional networks (VAN, DAN) explain differences in fluid and crystallized abilities. We found that stronger between-network connectivity of the DMN and VAN, DMN and DAN, and VAN and DAN predicted crystallized abilities. Within-network connectivity of the DAN predicted both crystallized and fluid abilities. Our findings reveal that crystallized abilities rely on the functional coupling between attentional networks and the DMN, whereas fluid abilities are associated with a focal connectivity configuration at the DAN. Our study provides new evidence into the neural basis of child intelligence and calls for future comparative research in adulthood during neuropsychiatric diseases.

Predictive overfitting in immunological applications: Pitfalls and solutions

Overfitting describes the phenomenon where a highly predictive model on the training data generalizes poorly to future observations. It is a common concern when applying machine learning techniques to contemporary medical applications, such as predicting vaccination response and disease status in infectious disease or cancer studies. This review examines the causes of overfitting and offers strategies to counteract it, focusing on model complexity reduction, reliable model evaluation, and harnessing data diversity. Through discussion of the underlying mathematical models and illustrative examples using both synthetic data and published real datasets, our objective is to equip analysts and bioinformaticians with the knowledge and tools necessary to detect and mitigate overfitting in their research.

Decoding numeracy and literacy in the human brain: insights from MEG and MVPA

Numbers and letters are the fundamental building blocks of our everyday social interactions. Previous studies have focused on determining the cortical pathways shaped by numeracy and literacy in the human brain, partially supporting the hypothesis of distinct perceptual neural circuits involved in the visual processing of the two categories. In this study, we aim to investigate the temporal dynamics for number and letter processing. We present magnetoencephalography (MEG) data from two experiments (N = 25 each). In the first experiment, single numbers, letters, and their respective false fonts (false numbers and false letters) were presented, whereas, in the second experiment, numbers, letters, and their respective false fonts were presented as a string of characters. We used multivariate pattern analysis techniques (time-resolved decoding and temporal generalization), testing the strong hypothesis that the neural correlates supporting letter and number processing can be logistically classified as categorically separate. Our results show a very early dissociation (~ 100 ms) between numbers, and letters when compared to false fonts. Number processing can be dissociated with similar accuracy when presented as isolated items or strings of characters, while letter processing shows dissociable classification accuracy for single items compared to strings. These findings reinforce the evidence indicating that early visual processing can be differently shaped by the experience with numbers and letters; this dissociation is stronger for strings compared to single items, thus showing that combinatorial mechanisms for numbers and letters could be categorically distinguished and influence early visual processing.

Sampling inequalities affect generalization of neuroimaging-based diagnostic classifiers in psychiatry

BACKGROUND

The development of machine learning models for aiding in the diagnosis of mental disorder is recognized as a significant breakthrough in the field of psychiatry. However, clinical practice of such models remains a challenge, with poor generalizability being a major limitation.

METHODS

Here, we conducted a pre-registered meta-research assessment on neuroimaging-based models in the psychiatric literature, quantitatively examining global and regional sampling issues over recent decades, from a view that has been relatively underexplored. A total of 476 studies (n = 118,137) were included in the current assessment. Based on these findings, we built a comprehensive 5-star rating system to quantitatively evaluate the quality of existing machine learning models for psychiatric diagnoses.

RESULTS

A global sampling inequality in these models was revealed quantitatively (sampling Gini coefficient (G) = 0.81, p < .01), varying across different countries (regions) (e.g., China, G = 0.47; the USA, G = 0.58; Germany, G = 0.78; the UK, G = 0.87). Furthermore, the severity of this sampling inequality was significantly predicted by national economic levels (β =  - 2.75, p < .001, R²_adj = 0.40; r =  - .84, 95% CI: - .41 to - .97), and was plausibly predictable for model performance, with higher sampling inequality for reporting higher classification accuracy. Further analyses showed that lack of independent testing (84.24% of models, 95% CI: 81.0-87.5%), improper cross-validation (51.68% of models, 95% CI: 47.2-56.2%), and poor technical transparency (87.8% of models, 95% CI: 84.9-90.8%)/availability (80.88% of models, 95% CI: 77.3-84.4%) are prevailing in current diagnostic classifiers despite improvements over time. Relating to these observations, model performances were found decreased in studies with independent cross-country sampling validations (all p < .001, BF₁₀ > 15). In light of this, we proposed a purpose-built quantitative assessment checklist, which demonstrated that the overall ratings of these models increased by publication year but were negatively associated with model performance.

CONCLUSIONS

Together, improving sampling economic equality and hence the quality of machine learning models may be a crucial facet to plausibly translating neuroimaging-based diagnostic classifiers into clinical practice.

A Guide to Cross-Validation for Artificial Intelligence in Medical Imaging

Artificial intelligence (AI) is being increasingly used to automate and improve technologies within the field of medical imaging. A critical step in the development of an AI algorithm is estimating its prediction error through cross-validation (CV). The use of CV can help prevent overoptimism in AI algorithms and can mitigate certain biases associated with hyperparameter tuning and algorithm selection. This article introduces the principles of CV and provides a practical guide on the use of CV for AI algorithm development in medical imaging. Different CV techniques are described, as well as their advantages and disadvantages under different scenarios. Common pitfalls in prediction error estimation and guidance on how to avoid them are also discussed. Keywords: Education, Research Design, Technical Aspects, Statistics, Supervised Learning, Convolutional Neural Network (CNN) Supplemental material is available for this article. © RSNA, 2023.

Aberrant topology of white matter networks in patients with methamphetamine dependence and its application in support vector machine-based classification

Brain white matter (WM) networks have been widely studied in neuropsychiatric disorders. However, few studies have evaluated alterations in WM network topological organization in patients with methamphetamine (MA) dependence. Therefore, using machine learning classification methods to analyze WM network topological attributes may give new insights into patients with MA dependence. In the study, diffusion tensor imaging-based probabilistic tractography was used to map the weighted WM networks in 46 MA-dependent patients and 46 control subjects. Using graph-theoretical analyses, the global and regional topological attributes of WM networks for both groups were calculated and compared to determine inter-group differences using a permutation-based general linear model. In addition, the study used a support vector machine (SVM) learning approach to construct a classifier for discriminating subjects with MA dependence from control subjects. Relative to the control group, the MA-dependent group exhibited abnormal topological organization, as evidenced by decreased small-worldness and modularity, and increased nodal efficiency in the right medial superior temporal gyrus, right pallidum, and right ventromedial putamen; the MA-dependent group had the higher hubness scores in 25 regions, which were mainly located in the default mode network. An SVM trained with topological attributes achieved classification accuracy, sensitivity, specificity, and kappa values of 98.09% ± 2.59%, 98.24% ± 4.00%, 97.94% ± 4.26%, and 96.18% ± 5.19% for patients with MA dependence. Our results may suggest altered global WM structural networks in MA-dependent patients. Furthermore, the abnormal WM network topological attributes may provide promising features for the construction of high-efficacy classification models.

Brain-age prediction: A systematic comparison of machine learning workflows

The difference between age predicted using anatomical brain scans and chronological age, i.e., the brain-age delta, provides a proxy for atypical aging. Various data representations and machine learning (ML) algorithms have been used for brain-age estimation. However, how these choices compare on performance criteria important for real-world applications, such as; (1) within-dataset accuracy, (2) cross-dataset generalization, (3) test-retest reliability, and (4) longitudinal consistency, remains uncharacterized. We evaluated 128 workflows consisting of 16 feature representations derived from gray matter (GM) images and eight ML algorithms with diverse inductive biases. Using four large neuroimaging databases covering the adult lifespan (total N = 2953, 18-88 years), we followed a systematic model selection procedure by sequentially applying stringent criteria. The 128 workflows showed a within-dataset mean absolute error (MAE) between 4.73-8.38 years, from which 32 broadly sampled workflows showed a cross-dataset MAE between 5.23-8.98 years. The test-retest reliability and longitudinal consistency of the top 10 workflows were comparable. The choice of feature representation and the ML algorithm both affected the performance. Specifically, voxel-wise feature spaces (smoothed and resampled), with and without principal components analysis, with non-linear and kernel-based ML algorithms performed well. Strikingly, the correlation of brain-age delta with behavioral measures disagreed between within-dataset and cross-dataset predictions. Application of the best-performing workflow on the ADNI sample showed a significantly higher brain-age delta in Alzheimer's and mild cognitive impairment patients compared to healthy controls. However, in the presence of age bias, the delta estimates in the patients varied depending on the sample used for bias correction. Taken together, brain-age shows promise, but further evaluation and improvements are needed for its real-world application.