Biology subjects

Calhoun, V. D.

Publications and source records attributed to Calhoun, V. D..

At least 37 records · Page 2Linked to original sources

bioRxiv2025

aNy-way ICA and its application to estimate cortico-thalamo-cerebellar functional links in schizophrenia

Disrupted large-scale brain network interactions are a hallmark of schizophrenia, yet the circuit-level mechanisms underlying cognitive deficits remain incompletely understood. In particular, cortico-thalamo-cerebellar circuitry has been implicated in "cognitive dysmetria," but its functional organization across subdomains has not been fully characterized. Here, we apply a flexible multimodal data fusion framework (aNy-way independent component analysis) to integrate 4D domain-specific resting-state fMRI data from the cortex, thalamus, and cerebellum. Using a large discovery cohort (FBIRN, N=311) and two independent replication datasets (COBRE, N=148; MPRC, N=364), we identify a robust cortico-thalamo-cerebellar circuit involving higher-order thalamic nuclei, visual cortex, default mode network, and cerebellar posterior regions. This circuit shows altered connectivity in schizophrenia, characterized by reduced anticorrelation between thalamus and cortex, and is consistently associated with cognitive impairments, including processing speed, working memory, and reasoning deficits across cohorts. Importantly, these findings are largely replicated across independent cohorts. Together, these results provide data-driven evidence for a cortico-thalamo-cerebellar mechanism underlying cognitive dysfunction in schizophrenia, supporting the cognitive dysmetria hypothesis. This work also demonstrates the utility of aNy-way independent component analysis for identifying clinically relevant brain circuits in psychiatric disorders. HighlightsO_LIIdentified cortico-thalamo-cerebellar circuits in schizophrenia C_LIO_LIAltered thalamocortical interactions differentiated schizophrenia C_LIO_LICircuit alterations were associated with cognitive impairment C_LIO_LIFindings are replicated across three independent multisite datasets C_LIO_LIMultidomain fusion enabled subdomain-specific circuit characterization C_LI

neuroscience↗

bioRxiv2025

Cellular and molecular underpinnings of functional networks in the human brain

Understanding how cellular and molecular architecture underpins the large-scale organization of human brain function is a central challenge in neuroscience. By integrating transcriptomic (microarray and single-nucleus RNA-sequencing), molecular imaging, and neuroimaging datasets, we observe spatial correspondences indicating that the distributions of diverse cell types, neurotransmitter systems, and mitochondrial phenotypes align with intrinsic connectivity networks (ICNs). These associations extend beyond local correspondence to reflect network-level structure: inter-ICN similarity networks derived from cellular and molecular profiles recapitulate static and dynamic patterns of functional network connectivity (FNC), mirroring canonical functional domains. Mediation analyses reveal that specific ICNs mediate the relationship between microscale cell-type architecture and domain-specific cognitive processes, while FNCs capture mediating pathways linking cell-type and neurotransmitter similarity networks to cognitive organization. Together, our findings show that the brains functional architecture systematically aligns with cellular and molecular organization, which may constrain functional network formation and contribute to the neural basis of cognition.

neuroscience↗

bioRxiv2025

Spatiotemporal Complexity in the Psychotic Brain

Psychotic disorders, such as schizophrenia and bipolar disorder, pose significant diagnostic challenges with major implications on mental health. The measures of resting-state fMRI spatiotemporal complexity offer a powerful tool for identifying irregularities in brain activity. To capture global brain connectivity, we employed information-theoretic metrics, overcoming the limitations of pairwise correlation analysis approaches. This enables a more comprehensive exploration of higher-order interactions and multiscale intrinsic connectivity networks (ICNs) in the psychotic brain. In this study, we provide converging evidence suggesting that the psychotic brain exhibits states of randomness across both spatial and temporal dimensions. To further investigate these disruptions, we estimated brain network connectivity using redundancy and synergy measures, aiming to assess the integration and segregation of topological information in the psychotic brain. Our findings reveal a disruption in the balance between redundant and synergistic information, a phenomenon we term brainquake in this study, which highlights the instability and disorganization of brain networks in psychosis. Moreover, our exploration of higher-order topological functional connectivity reveals profound disruptions in brain information integration. Aberrant information interactions were observed across both cortical and subcortical ICNs. We specifically identified the most easily affected irregularities in the sensorimotor, visual, temporal, default mode, and fronto-parietal networks, as well as in the hippocampal and amygdalar regions, all of which showed disruptions. These findings underscore the severe impact of psychotic states on multiscale critical brain networks, suggesting a profound alteration in the brains complexity and organizational states.

neuroscience↗

bioRxiv2024

Visualizing Functional Network Connectivity Differences Using an Explainable Machine-learning Method

Functional network connectivity (FNC) estimated from resting-state functional magnetic resonance imaging showed great information about the neural mechanism in different brain disorders. But previous research has mainly focused on standard statistical learning approaches to find FNC features separating patients from control. Although machine learning approaches provide better models separating controls from patients, it is not straightforward for these approaches to provide intuition on the model and the underlying neural process of each disorder. Explainable machine learning offers a solution to this problem by applying machine learning to understand the neural process behind brain disorders. In this study, we introduce a novel framework leveraging SHapley Additive exPlanations (SHAP) to identify crucial Functional Network Connectivity (FNC) features distinguishing between two distinct population classes. Initially, we validate our approach using synthetic data. Subsequently, applying our framework, we ascertain FNC biomarkers distinguishing between, controls and schizophrenia patients with accuracy of 81.04% as well as middle aged adults and old aged adults with accuracy 71.38%, respectively, employing Random Forest (RF), XGBoost, and CATBoost models. Our analysis underscores the pivotal role of the cognitive control network (CCN), subcortical network (SCN), and somatomotor network (SMN) in discerning individuals with schizophrenia from controls. In addition, our platform found CCN and SCN as the most important networks separating young adults from older.

neuroscience↗

bioRxiv2024

Image Processing in the Acute to Chronic Pain Signatures (A2CPS) Project

The Acute to Chronic Pain Signatures (A2CPS) project is a large-scale, multi-site initiative aimed at identifying biomarkers and biosignatures that predict the transition from acute to chronic pain. The project is collecting multimodal, longitudinal data from over 2,500 individuals at risk for developing chronic pain after surgery. Here we describe the neuroimaging component of A2CPS, including the acquisition protocols, processing pipelines, and contents of the initial data release. The imaging protocol includes structural, diffusion, resting-state and task-based functional magnetic resonance imaging (MRI) data. Data are collected across multiple clinical sites using different scanner manufacturers, with attention to protocol harmonization and quality control. The processing pipeline integrates several established neuroimaging tools to extract potential biomarkers, including measures of brain structure, connectivity, and pain-related neural signatures. The first data release includes pre-surgical imaging data for 595 participants, with high quality ratings across modalities (98.7% of sMRI, 99.8% of dMRI, and 94.6% of fMRI images were rated as acceptable or better). Initial analyses demonstrate expected relationships between brain-derived measures and clinical variables, such as associations between brain age and psychological factors. This dataset represents a valuable resource for both pain research and neuroimaging methods development, with future releases planned to include additional participants and expanded analysis pipelines and processed data derivatives.

neuroscience↗

bioRxiv2024

Evidence for white matter intrinsic connectivity networks at rest and during a task: a large-scale study and templates

Understanding white matter (WM) functional connectivity is crucial for unraveling brain function and dysfunction. In this study, we present a novel WM intrinsic connectivity network (ICN) template derived from over 100,000 fMRI scans, identifying 97 robust WM ICNs using spatially constrained independent component analysis (scICA). This WM template, combined with a previously identified gray matter (GM) ICN template from the same dataset, was applied to analyze a resting-state fMRI (rs-fMRI) dataset from the Bipolar-Schizophrenia Network on Intermediate Phenotypes 2 (BSNIP2; 590 subjects) and a task-based fMRI dataset from the MIND Clinical Imaging Consortium (MCIC; 75 subjects). Our analysis highlights distinct spatial maps for WM and GM ICNs, with WM ICNs showing higher frequency profiles. Modular structure within WM ICNs and interactions between WM and GM modules were identified. Task-based fMRI revealed event-related BOLD signals in WM ICNs, particularly within the corticospinal tract, lateralized to finger movement. Notable differences in static functional network connectivity (sFNC) matrices were observed between controls (HC) and schizophrenia (SZ) subjects in both WM and GM networks. This open-source WM NeuroMark template and automated pipeline offer a powerful tool for advancing WM connectivity research across diverse datasets.

neuroscience↗

bioRxiv2024

Investigating the Impact of Habitual Sleep Quality on Episodic Memory Performance: An EEG-Based Representational Similarity Analysis

Sleep is crucial for episodic memory consolidation, yet the impact of habitual sleep quality on memory performance remains underexplored. This study investigates the relationship between sleep quality and episodic memory retrieval using EEG-based representational similarity analysis (RSA). Thirty-six participants wore wrist accelerometers for one week to capture habitual sleep patterns, including total sleep time and restlessness. Memory performance was assessed through a paired associate learning task, with EEG data recorded during encoding and retrieval phases. RSA was applied to EEG oscillatory power across time-frequency windows to examine the neural similarity between encoding and retrieval. The results showed both positive and negative correlations between sleep metric and memory performance, with sleep restlessness being linked to both increases and decreases in neural similarity across specific clusters. These findings emphasize the important role of sleep quality in shaping the neural processes underlying episodic memory retrieval, indicating a strong connection between sleep patterns and memory function.

neuroscience↗

bioRxiv2024

A Method of Feature Selection via Deep Convolution Neural Networks For Encoding Nonlinear Functional Network Connectivity and Its Application To The Classification of Mental Disorders

In functional magnetic resonance imaging (fMRI) studies, it is common to evaluate the brains functional network connectivity (FNC) which captures the temporal coupling between hemodynamic signals derived from whole brain networks. FNC has been linked to various psychological phenomena. However, analysis of FNCs mainly focuses on linear statistical relationships, which may not capture the full complexity of the interactions among brain intrinsic connectivity networks (ICNs). Therefore, it is important to explore approaches that can better account for possible intricate nonlinear interactions involved in cognitive operations and the changes observed in psychiatric conditions such as schizophrenia. This exploration can lead to a better understanding of brain function and provide new insights into neural links to various psychological and psychiatric conditions. In this paper, we present an innovative approach which utilizes a deep convolutional neural network (DCNN) to extract nonlinear heatmaps from FNC matrices. By analyzing the heatmaps, multi-level nonlinear interactions can be derived from the corresponding input FNC data. Our results show these networks represent a significant improvement over previous approaches and offer a robust framework for understanding the complex interactions between brain regions. By incorporating two stages in the training process, our method ensures optimal efficiency and effectiveness. In the initial stage, a deep convolutional neural network is trained to create heatmaps from various convolution layers of the network. In the next stage, by utilizing a t-test-based feature selection method, we can effectively analyze heatmaps from different convolution layers. This approach ensures that we are able to functional connectivity with varying degrees of nonlinearity with a focus on the heatmaps that play an important role in distinguishing different groups. We used a large dataset consisting of both schizophrenia patients and healthy controls, which were divided into separate training and validation sets to evaluate this approach. Results showed patients with increases in default mode networks connections to itself and cognitive control regions and controls with increases within visual and between visual, motor, and auditory domains. We also find significantly increased cross-validated classification accuracy (at 92.8%) compared to several competing approaches. Our approach shows the potential to accurately distinguish differences between the schizophrenia and healthy control groups with high accuracy.

neuroscience↗

bioRxiv2024

Spatial Development of Brain Networks During The First Six Postnatal Months

The initial months of life constitute a crucial period for human development. A comprehensive understanding of this early phase is essential for unraveling the origins of neurodevelopmental disorders and promoting infant brain health. This study uniquely focuses on the spatial development of intrinsic brain connectivity networks during infancy, which has been less explored compared to functional connectivity. We utilized independent component analysis on resting-state fMRI data from 74 infants to assess how the spatial organization of infant brain networks evolves between birth and six months. Our findings reveal significant changes in spatial characteristics, including an a notable rise in the network-averaged spatial similarity across age, reflecting how closely each participant-specific spatial map aligns with the group-level map for each network. We also observed a marked reduction in the network engagement range by age, representing the extent of voxel intensity range fluctuation within each network. This suggests a continuing process of consolidation, where voxel contributions to the network become more uniform, as indicated by the narrowing of intensity values. The network strength, calculated as the average of all the voxel intensities in the network, indicating the degree of involvement to the specific functional network, increased across age in several networks, such as frontal-mPFC, primary, and secondary visual networks. The network size, along with the network center of mass, illustrating spatial distribution alterations of brain networks by age, varied across different networks. For instance, both metrics increased across age in the secondary visual network but decreased in the temporal network. Additionally, we examined the networks in relation to their linear versus non-linear developmental trajectories across all spatial characteristics, providing a deeper understanding of how these patterns evolve during early infancy. These findings contribute to early brain development understanding and offer insights into potential markers of consolidation and spatial reorganization in large-scale brain networks during infancy.

neuroscience↗

bioRxiv2024

Addressing inconsistency in functional neuroimaging: A replicable data-driven multi-scale functional atlas for canonical brain networks

The advent of multiple neuroimaging methodologies has greatly aided in the conceptualization of large-scale functional brain networks in the field of cognitive neuroscience. However, there is inconsistency across studies in both nomenclature and the functional entities being described. There is a need for a unifying framework that standardizes terminology across studies while also bringing analyses and results into the same reference space. Here we present a whole-brain atlas of canonical functional brain networks derived from more than 100,000 resting-state fMRI datasets. These data-driven functional networks are highly replicable across datasets and capture information from multiple spatial scales. We have organized, labeled, and described the networks with terms familiar to the fields of cognitive and affective neuroscience in order to optimize their utility in future neuroimaging analyses and enhance the accessibility of new findings. The benefits of this atlas are not limited to future template-based or reference-guided analyses, but also extend to other data-driven neuroimaging approaches across modalities, such as those using blind independent component analysis (ICA). Future studies utilizing this atlas will contribute to greater harmonization and standardization in functional neuroimaging research.

neuroscience↗

bioRxiv2024

Static and Dynamic Cross-Network Functional Connectivity Shows Elevated Entropy in Schizophrenia Patients

Schizophrenia (SZ) patients exhibit abnormal static and dynamic functional connectivity across various brain domains. We present a novel approach based on static and dynamic inter-network connectivity entropy (ICE), which represents the entropy of a given networks connectivity to all the other brain networks. This novel approach enables the investigation of how connectivity strength is heterogeneously distributed across available targets in both SZ patients and healthy controls. We analyzed fMRI data from 151 schizophrenia patients and demographically matched 160 healthy controls. Our assessment encompassed both static and dynamic ICE, revealing significant differences in the heterogeneity of connectivity levels across available brain networks between SZ patients and healthy controls (HC). These networks are associated with subcortical (SC), auditory (AUD), sensorimotor (SM), visual (VIS), cognitive control (CC), default mode network (DMN) and cerebellar (CB) functional brain domains. Elevated ICE observed in individuals with SZ suggests that patients exhibit significantly higher randomness in the distribution of time-varying connectivity strength across functional regions from each source network, compared to healthy control group. C-means fuzzy clustering analysis of functional ICE correlation matrices revealed that SZ patients exhibit significantly higher occupancy weights in clusters with weak, low-scale functional entropy correlation, while the control group shows greater occupancy weights in clusters with strong, large-scale functional entropy correlation. k-means clustering analysis on time-indexed ICE vectors revealed that cluster with highest ICE have higher occupancy rates in SZ patients whereas clusters characterized by lowest ICE have larger occupancy rates for control group. Furthermore, our dynamic ICE approach revealed that it appears healthy for a brain to primarily circulate through complex, less structured connectivity patterns, with occasional transitions into more focused patterns. However, individuals with SZ seem to struggle with transiently attaining these more focused and structured connectivity patterns. Proposed ICE measure presents a novel framework for gaining deeper insights into understanding mechanisms of healthy and disease brain states and a substantial step forward in the developing advanced methods of diagnostics of mental health conditions.

neuroscience↗

bioRxiv2024

Anterior pituitary gland volume mediates associations between pubertal hormones and changes in transdiagnostic symptoms in youth

The pituitary gland (PG) plays a central role in the production and secretion of pubertal hormones, with documented links to the emergence and increase in mental health symptoms known to occur during adolescence. Although much of the literature has focused on examining whole PG volume, recent findings suggest that there are associations among pubertal hormone levels, including dehydroepiandrosterone (DHEA), subregions of the PG, and elevated mental health symptoms (e.g., internalizing symptoms) during adolescence. Surprisingly, studies have not yet examined associations among these factors and increasing transdiagnostic symptomology, despite DHEA being a primary output of the anterior PG. Therefore, the current study sought to fill this gap by examining whether anterior PG volume specifically mediates associations between DHEA levels and changes in dysregulation symptoms in an adolescent sample (N = 114, 9 - 17 years, Mage = 12.87, SD = 1.88). Following manual tracing of the anterior and posterior PG, structural equation modeling revealed that greater anterior, not posterior, PG volume mediated the association between greater DHEA levels and increasing dysregulation symptoms across time, controlling for baseline dysregulation symptom levels. These results suggest specificity in the role of the anterior PG in adrenarcheal processes that may confer risk for psychopathology during adolescence. This work not only highlights the importance of separately tracing the anterior and posterior PG, but also suggests that transdiagnostic factors like dysregulation are useful in parsing hormone-related increases in mental health symptoms in youth.

neuroscience↗

bioRxiv2024

A Dynamic Entropy Approach Reveals Reduced Functional Network Connectivity Trajectory Complexity in Schizophrenia

Over the past decade and a half, dynamic functional imaging has revolutionized the neuroimaging field. Since 2009, it has revealed low dimensional brain connectivity measures, has identified potential common human spatial connectivity states, has tracked the transition patterns of these states, and has demonstrated meaningful alterations in these transition and spatial patterns in neurological disorders, psychiatric disorders, and over the course of development. More recently, researchers have begun to analyze this data from the perspective of dynamic system and information theory in the hopes of understanding the constraints within which these dynamics occur and how they may support less easily quantified processes, such as information processing, cortical hierarchy, and consciousness. Progress has begun to accelerate in this area, particularly around consciousness, which appears to be strongly linked to entropy production in the brain. Outside of disorders of consciousness, however, little attention has been paid to the effects of psychiatric disease on entropy production in the human brain. Even disorders characterized by substantial changes in conscious experience have not been widely analyzed from this perspective. Here, we begin to rectify this gap by examining the complexity of subject trajectories in this state space through the lens of information theory. Specifically, we identify a basis for the dynamic functional connectivity state space and track subject trajectories through this state space over the course of the scan. The dynamic complexity of these trajectories is estimated using a Kozachenko-Leonenko entropy estimator, which assesses the rate of Shannon entropy production along each dimension of the proposed basis space. Using these estimates, we demonstrate that schizophrenia patients display substantially simpler trajectories than demographically matched healthy controls, and that this drop in complexity concentrates along specific dimensions of projected basis space. We also demonstrate that entropy generation in at least one of these dimensions is linked to cognitive performance. Overall, results suggest great value in applying dynamic systems theory to problems of neuroimaging and reveal a substantial drop in the complexity of schizophrenia patients brain function.

neuroscience↗

bioRxiv2024

Uncovering Effects of Schizophrenia upon a Maximally Significant, Minimally Complex Subset of Default Mode Network Connectivity Features

A common analysis approach for resting state functional magnetic resonance imaging (rs-fMRI) dynamic functional network connectivity (dFNC) data involves clustering windowed correlation time-series and assigning time windows to clusters (i.e., states) that can be quantified to summarize aspects of the dFNC dynamics. However, those methods can be dominated by a select few features and obscure key dynamics related to less dominant features. This study presents an iterative feature learning approach to identify a maximally significant and minimally complex subset of dFNC features within the default mode network (DMN) in schizophrenia (SZ). Utilizing dFNC data from individuals with SZ and healthy controls (HC), our approach uncovers a subset of features that has a greater number of dFNC states with disorder-related dynamics than is found when all features are present in the clustering. We find that anterior cingulate cortex/posterior cingulate cortex (ACC/PCC) interactions are consistently related to SZ across the most significant iterations of the feature learning analysis and that individuals with SZ tend to spend more time in states with greater intra-ACC anticorrelation and almost no time in a state of high intra-ACC correlation that HCs periodically enter. Our findings highlight the need for nuanced analyses to reveal disorder-related dynamics and advance our understanding of neuropsychiatric disorders.

neuroscience↗

bioRxiv2024

Parallel Multilink Group Joint ICA: Fusion of 3D Structural and 4D Functional Data Across Multiple Resting fMRI Networks

Multimodal neuroimaging research plays a pivotal role in understanding the complexities of the human brain and its disorders. Independent component analysis (ICA) has emerged as a widely used and powerful tool for disentangling mixed independent sources, particularly in the analysis of functional magnetic resonance imaging (fMRI) data. This paper extends the use of ICA as a unifying framework for multimodal fusion, introducing a novel approach termed parallel multilink group joint ICA (pmg-jICA). The method allows for the fusion of gray matter maps from structural MRI (sMRI) data to multiple fMRI intrinsic networks, addressing the limitations of previous models. The effectiveness of pmg-jICA is demonstrated through its application to an Alzheimers dataset, yielding linked structure-function outputs for 53 brain networks. Our approach leverages the complementary information from various imaging modalities, providing a unique perspective on brain alterations in Alzheimers disease. The pmg-jICA identifies several components with significant differences between HC and AD groups including thalamus, caudate, putamen with in the subcortical (SC) domain, insula, parahippocampal gyrus within the cognitive control (CC) domain, and the lingual gyrus within the visual (VS) domain, providing localized insights into the links between AD and specific brain regions. In addition, because we link across multiple brain networks, we can also compute functional network connectivity (FNC) from spatial maps and subject loadings, providing a detailed exploration of the relationships between different brain regions and allowing us to visualize spatial patterns and loading parameters in sMRI along with intrinsic networks and FNC from the fMRI data. In essence, developed approach combines concepts from joint ICA and group ICA to provide a rich set of output characterizing data-driven links between covarying gray matter networks, and a (potentially large number of) resting fMRI networks allowing further study in the context of structure/function links. We demonstrate the utility of the approach by highlighting key structure/function disruptions in Alzheimers individuals.

neuroscience↗

bioRxiv2024

Identifying EEG Biomarkers of Depression with Novel Explainable Deep Learning Architectures

Deep learning methods are increasingly being applied to raw electro-encephalogram (EEG) data. However, if these models are to be used in clinical or research contexts, methods to explain them must be developed, and if these models are to be used in research contexts, methods for combining explanations across large numbers of models must be developed to counteract the inherent randomness of existing training approaches. Model visualization-based explainability methods for EEG involve structuring a model architecture such that its extracted features can be characterized and have the potential to offer highly useful insights into the patterns that they uncover. Nevertheless, model visualization-based explainability methods have been underexplored within the context of multichannel EEG, and methods to combine their explanations across folds have not yet been developed. In this study, we present two novel convolutional neural network-based architectures and apply them for automated major depressive disorder diagnosis. Our models obtain slightly lower classification performance than a baseline architecture. However, across 50 training folds, they find that individuals with MDD exhibit higher {beta} power, potentially higher {delta} power, and higher brain-wide correlation that is most strongly represented within the right hemisphere. This study provides multiple key insights into MDD and represents a significant step forward for the domain of explainable deep learning applied to raw EEG. We hope that it will inspire future efforts that will eventually enable the development of explainable EEG deep learning models that can contribute both to clinical care and novel medical research discoveries.

neuroscience↗

bioRxiv2024

Beyond functional connectivity: deep learning applied to resting-state fMRI time series in the prediction of 58 human traits in the HCP

Machine learning has made several inroads into the study of brain-behavior relations based on in vivo imaging. While the advent of deep neural networks was expected to further improve predictions, the current literature based on resting-state functional connectivity presents mixed results. We hypothesize that the representation of the data, i.e. in the form of functional connectivity, could restrict an advantage of deep learning techniques, namely that of learning complex representations directly from the data. Thus, we investigated if bypassing this feature extraction resulted in improved performance in the prediction of 58 widely studied behavioral traits from a large sample of Human Connectome Project subjects, using deep learning techniques. For this task, we adapted the InceptionTime architecture, which jointly predicts traits directly from regional time series through representation learning, and compared results with a strong kernel-based baseline. Results revealed that both models achieve comparable performance in most traits. Eleven significant differences in mean squared error were detected, however, with seven favoring the neural network approach, and this number increased when accounting for covariates. We additionally show that contrary to the expectation, the neural network approach was more robust to reductions in the training set size. On the other hand, it was more sensitive to reductions in the length of the time series at test time. Our results present a more nuanced view of the potential of deep learning for the prediction of behavior from neuroimaging, which allows learning features directly from the data.

neuroscience↗

bioRxiv2024

Identifying Reproducibly Important EEG Markers of Schizophrenia with an Explainable Multi-Model Deep Learning Approach

The diagnosis of schizophrenia (SZ) can be challenging due to its diverse symptom presentation. As such, many studies have sought to identify diagnostic biomarkers of SZ using explainable machine learning methods. However, the generalizability of identified biomarkers in many machine learning-based studies is highly questionable given that most studies only analyze explanations from a small number of models. In this study, we present (1) a novel feature interaction-based explainability approach and (2) several new approaches for summarizing multi-model explanations. We implement our approach within the context of electroencephalogram (EEG) spectral power data. We further analyze both training and test set explanations with the goal of extracting generalizable insights from the models. Importantly, our analyses identify effects of SZ upon the , {beta}, and {theta} frequency bands, the left hemisphere of the brain, and interhemispheric interactions across a majority of folds. We hope that our analysis will provide helpful insights into SZ and inspire the development of robust approaches for identifying neuropsychiatric disorder biomarkers from explainable machine learning models.

neuroscience↗