Sex classification from functional brain connectivity: Generalization to multiple datasets.

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

Network and state specificity in connectivity-based predictions of individual behavior.

Predicting individual behavior from brain functional connectivity (FC) patterns can contribute to our understanding of human brain functioning. This may apply in particular if predictions are based on features derived from circumscribed, a priori defined functional networks, which improves interpretability. Furthermore, some evidence suggests that task-based FC data may yield more successful predictions of behavior than resting-state FC data. Here, we comprehensively examined to what extent the correspondence of functional network priors and task states with behavioral target domains influences the predictability of individual performance in cognitive, social, and affective tasks. To this end, we used data from the Human Connectome Project for large-scale out-of-sample predictions of individual abilities in working memory (WM), theory-of-mind cognition (SOCIAL), and emotion processing (EMO) from FC of corresponding and non-corresponding states (WM/SOCIAL/EMO/resting-state) and networks (WM/SOCIAL/EMO/whole-brain connectome). Using root mean squared error and coefficient of determination to evaluate model fit revealed that predictive performance was rather poor overall. Predictions from whole-brain FC were slightly better than those from FC in task-specific networks, and a slight benefit of predictions based on FC from task versus resting state was observed for performance in the WM domain. Beyond that, we did not find any significant effects of a correspondence of network, task state, and performance domains. Together, these results suggest that multivariate FC patterns during both task and resting states contain rather little information on individual performance levels, calling for a reconsideration of how the brain mediates individual differences in mental abilities.

Predicting executive functioning from functional brain connectivity: network specificity and age effects.

Healthy aging is associated with altered executive functioning (EF). Earlier studies found age-related differences in EF performance to be partially accounted for by changes in resting-state functional connectivity (RSFC) within brain networks associated with EF. However, it remains unclear which role RSFC in EF-associated networks plays as a marker for individual differences in EF performance. Here, we investigated to what degree individual abilities across 3 different EF tasks can be predicted from RSFC within EF-related, perceptuo-motor, whole-brain, and random networks separately in young and old adults. Specifically, we were interested if (i) young and old adults differ in predictability depending on network or EF demand level (high vs. low), (ii) an EF-related network outperforms EF-unspecific networks when predicting EF abilities, and (iii) this pattern changes with demand level. Both our uni- and multivariate analysis frameworks analyzing interactions between age × demand level × networks revealed overall low prediction accuracies and a general lack of specificity regarding neurobiological networks for predicting EF abilities. This questions the idea of finding markers for individual EF performance in RSFC patterns and calls for future research replicating the current approach in different task states, brain modalities, different, larger samples, and with more comprehensive behavioral measures.

Predicting executive functioning from brain networks: modality specificity and age effects.

Healthy aging is associated with structural and functional network changes in the brain, which have been linked to deterioration in executive functioning (EF), while their neural implementation at the individual level remains unclear. As the biomarker potential of individual resting-state functional connectivity (RSFC) patterns has been questioned, we investigated to what degree individual EF abilities can be predicted from the gray-matter volume (GMV), regional homogeneity, fractional amplitude of low-frequency fluctuations (fALFF), and RSFC within EF-related, perceptuo-motor, and whole-brain networks in young and old adults. We examined whether the differences in out-of-sample prediction accuracy were modality-specific and depended on age or task-demand levels. Both uni- and multivariate analysis frameworks revealed overall low prediction accuracies and moderate-to-weak brain-behavior associations (R2 < 0.07, r < 0.28), further challenging the idea of finding meaningful markers for individual EF performance with the metrics used. Regional GMV, well linked to overall atrophy, carried the strongest information about individual EF differences in older adults, whereas fALFF, measuring functional variability, did so for younger adults. Our study calls for future research analyzing more global properties of the brain, different task-states and applying adaptive behavioral testing to result in sensitive predictors for young and older adults, respectively.

A topography-based predictive framework for naturalistic viewing fMRI.

Functional magnetic resonance imaging (fMRI) during naturalistic viewing (NV) provides exciting opportunities for studying brain functions in more ecologically valid settings. Understanding individual differences in brain functions during NV and their behavioural relevance has recently become an important goal. However, methods specifically designed for this purpose remain limited. Here, we propose a topography-based predictive framework (TOPF) to fill this methodological gap. TOPF identifies individual-specific evoked activity topographies in a data-driven manner and examines their behavioural relevance using a machine learning-based predictive framework. We validate TOPF on both NV and task-based fMRI data from multiple conditions. Our results show that TOPF effectively and stably captures individual differences in evoked brain activity and successfully predicts phenotypes across cognition, emotion and personality on unseen subjects from their activity topographies. Moreover, TOPF compares favourably with functional connectivity-based approaches in prediction performance, with the identified predictive brain regions being neurobiologically interpretable. Crucially, we highlight the importance of examining individual evoked brain activity topographies in advancing our understanding of the brain-behaviour relationship. We believe that the TOPF approach provides a simple but powerful tool for understanding brain-behaviour relationships on an individual level with a strong potential for clinical applications.

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.

A systematic comparison of VBM pipelines and their application to age prediction.

Voxel-based morphometry (VBM) analysis is commonly used for localized quantification of gray matter volume (GMV). Several alternatives exist to implement a VBM pipeline. However, how these alternatives compare and their utility in applications, such as the estimation of aging effects, remain largely unclear. This leaves researchers wondering which VBM pipeline they should use for their project. In this study, we took a user-centric perspective and systematically compared five VBM pipelines, together with registration to either a general or a study-specific template, utilizing three large datasets (n>500 each). Considering the known effect of aging on GMV, we first compared the pipelines in their ability of individual-level age prediction and found markedly varied results. To examine whether these results arise from systematic differences between the pipelines, we classified them based on their GMVs, resulting in near-perfect accuracy. To gain deeper insights, we examined the impact of different VBM steps using the region-wise similarity between pipelines. The results revealed marked differences, largely driven by segmentation and registration steps. We observed large variability in subject-identification accuracies, highlighting the interpipeline differences in individual-level quantification of GMV. As a biologically meaningful criterion we correlated regional GMV with age. The results were in line with the age-prediction analysis, and two pipelines, CAT and the combination of fMRIPrep for tissue characterization with FSL for registration, reflected age information better.

Naturalistic viewing increases individual identifiability based on connectivity within functional brain networks.

Naturalistic viewing (NV) is currently considered a promising paradigm for studying individual differences in functional brain organization. While whole brain functional connectivity (FC) under NV has been relatively well characterized, so far little work has been done on a network level. Here, we extend current knowledge by characterizing the influence of NV on FC in fourteen meta-analytically derived brain networks considering three different movie stimuli in comparison to resting-state (RS). We show that NV increases identifiability of individuals over RS based on functional connectivity in certain, but not all networks. Furthermore, movie stimuli including a narrative appear more distinct from RS. In addition, we assess individual variability in network FC by comparing within- and between-subject similarity during NV and RS. We show that NV can evoke individually distinct NFC patterns by increasing inter-subject variability while retaining within-subject similarity. Crucially, our results highlight that this effect is not observable across all networks, but rather dependent on the network-stimulus combination. Our results confirm that NV can improve the detection of individual differences over RS and underline the importance of selecting the appropriate combination of movie and cognitive network for the research question at hand.

Lack of structural brain alterations associated with insomnia: findings from the ENIGMA-Sleep Working Group.

Existing neuroimaging studies have reported divergent structural alterations in insomnia disorder (ID). In the present study, we performed a large-scale coordinated meta-analysis by pooling structural brain measures from 1085 subjects (mean [SD] age 50.5 [13.9] years, 50.2% female, 17.4% with insomnia) across three international Enhancing NeuroImaging Genetics through Meta-Analysis (ENIGMA)-Sleep cohorts. Two sites recruited patients with ID/controls: Freiburg (University of Freiburg Medical Center, Freiburg, Germany) 42/43 and KUMS (Kermanshah University of Medical Sciences, Kermanshah, Iran) 42/49, while the Study of Health in Pomerania (SHIP-Trend, University Medicine Greifswald, Greifswald, Germany) recruited population-based individuals with/without insomnia symptoms 75/662. The influence of insomnia on magnetic resonance imaging-based brain morphometry using an insomnia brain score was then assessed. Within each cohort, we used an ordinary least-squares linear regression to investigate the link between the individual regional cortical and subcortical volumes and the presence of insomnia symptoms. Then, we performed a fixed-effects meta-analysis across cohorts based on the first-level results. For the insomnia brain score, weighted logistic ridge regression was performed on one sample (Freiburg), which separated patients with ID from controls to train a model based on the segmentation measurements. Afterward, the insomnia brain scores were validated using the other two samples. The model was used to predict the log-odds of the subjects with insomnia given individual insomnia-related brain atrophy. After adjusting for multiple comparisons, we did not detect any significant associations between insomnia symptoms and cortical or subcortical volumes, nor could we identify a global insomnia-related brain atrophy pattern. Thus, we observed inconsistent brain morphology differences between individuals with and without insomnia across three independent cohorts. Further large-scale cross-sectional/longitudinal studies using both structural and functional neuroimaging are warranted to decipher the neurobiology of insomnia.

A too-good-to-be-true prior to reduce shortcut reliance.

Despite their impressive performance in object recognition and other tasks under standard testing conditions, deep networks often fail to generalize to out-of-distribution (o.o.d.) samples. One cause for this shortcoming is that modern architectures tend to rely on ǣshortcutsǥ superficial features that correlate with categories without capturing deeper invariants that hold across contexts. Real-world concepts often possess a complex structure that can vary superficially across contexts, which can make the most intuitive and promising solutions in one context not generalize to others. One potential way to improve o.o.d. generalization is to assume simple solutions are unlikely to be valid across contexts and avoid them, which we refer to as the too-good-to-be-true prior. A low-capacity network (LCN) with a shallow architecture should only be able to learn surface relationships, including shortcuts. We find that LCNs can serve as shortcut detectors. Furthermore, an LCN's predictions can be used in a two-stage approach to encourage a high-capacity network (HCN) to rely on deeper invariant features that should generalize broadly. In particular, items that the LCN can master are downweighted when training the HCN. Using a modified version of the CIFAR-10 dataset in which we introduced shortcuts, we found that the two-stage LCN-HCN approach reduced reliance on shortcuts and facilitated o.o.d. generalization.

A Connectivity-Based Psychometric Prediction Framework for Brain-Behavior Relationship Studies.

The recent availability of population-based studies with neuroimaging and behavioral measurements opens promising perspectives to investigate the relationships between interindividual variability in brain regions' connectivity and behavioral phenotypes. However, the multivariate nature of connectivity-based prediction model severely limits the insight into brain-behavior patterns for neuroscience. To address this issue, we propose a connectivity-based psychometric prediction framework based on individual regions' connectivity profiles. We first illustrate two main applications: 1) single brain region's predictive power for a range of psychometric variables and 2) single psychometric variable's predictive power variation across brain region. We compare the patterns of brain-behavior provided by these approaches to the brain-behavior relationships from activation approaches. Then, capitalizing on the increased transparency of our approach, we demonstrate how the influence of various data processing and analyses can directly influence the patterns of brain-behavior relationships, as well as the unique insight into brain-behavior relationships offered by this approach.

Functional parcellation of human and macaque striatum reveals human-specific connectivity in the dorsal caudate.

A wide homology between human and macaque striatum is often assumed as in both the striatum is involved in cognition, emotion and executive functions. However, differences in functional and structural organization between human and macaque striatum may reveal evolutionary divergence and shed light on human vulnerability to neuropsychiatric diseases. For instance, dopaminergic dysfunction of the human striatum is considered to be a pathophysiological underpinning of different disorders, such as Parkinson's disease (PD) and schizophrenia (SCZ). Previous investigations have found a wide similarity in structural connectivity of the striatum between human and macaque, leaving the cross-species comparison of its functional organization unknown. In this study, resting-state functional connectivity (RSFC) derived striatal parcels were compared based on their homologous cortico-striatal connectivity. The goal here was to identify striatal parcels whose connectivity is human-specific compared to macaque parcels. Functional parcellation revealed that the human striatum was split into dorsal, dorsomedial, and rostral caudate and ventral, central, and caudal putamen, while the macaque striatum was divided into dorsal, and rostral caudate and rostral, and caudal putamen. Cross-species comparison indicated dissimilar cortico-striatal RSFC of the topographically similar dorsal caudate. We probed clinical relevance of the striatal clusters by examining differences in their cortico-striatal RSFC and gray matter (GM) volume between patients (with PD and SCZ) and healthy controls. We found abnormal RSFC not only between dorsal caudate, but also between rostral caudate, ventral, central and caudal putamen and widespread cortical regions for both PD and SCZ patients. Also, we observed significant structural atrophy in rostral caudate, ventral and central putamen for both PD and SCZ while atrophy in the dorsal caudate was specific to PD. Taken together, our cross-species comparative results revealed shared and human-specific RSFC of different striatal clusters reinforcing the complex organization and function of the striatum. In addition, we provided a testable hypothesis that abnormalities in a region with human-specific connectivity, i.e., dorsal caudate, might be associated with neuropsychiatric disorders.

Imaging evolution of the primate brain: the next frontier?

Evolution, as we currently understand it, strikes a delicate balance between animals' ancestral history and adaptations to their current niche. Similarities between species are generally considered inherited from a common ancestor whereas observed differences are considered as more recent evolution. Hence comparing species can provide insights into the evolutionary history. Comparative neuroimaging has recently emerged as a novel subdiscipline, which uses magnetic resonance imaging (MRI) to identify similarities and differences in brain structure and function across species. Whereas invasive histological and molecular techniques are superior in spatial resolution, they are laborious, post-mortem, and oftentimes limited to specific species. Neuroimaging, by comparison, has the advantages of being applicable across species and allows for fast, whole-brain, repeatable, and multi-modal measurements of the structure and function in living brains and post-mortem tissue. In this review, we summarise the current state of the art in comparative anatomy and function of the brain and gather together the main scientific questions to be explored in the future of the fascinating new field of brain evolution derived from comparative neuroimaging.

Sex Classification by Resting State Brain Connectivity.

A large amount of brain imaging research has focused on group studies delineating differences between males and females with respect to both cognitive performance as well as structural and functional brain organization. To supplement existing findings, the present study employed a machine learning approach to assess how accurately participants' sex can be classified based on spatially specific resting state (RS) brain connectivity, using 2 samples from the Human Connectome Project (n1 = 434, n2 = 310) and 1 fully independent sample from the 1000BRAINS study (n = 941). The classifier, which was trained on 1 sample and tested on the other 2, was able to reliably classify sex, both within sample and across independent samples, differing both with respect to imaging parameters and sample characteristics. Brain regions displaying highest sex classification accuracies were mainly located along the cingulate cortex, medial and lateral frontal cortex, temporoparietal regions, insula, and precuneus. These areas were stable across samples and match well with previously described sex differences in functional brain organization. While our data show a clear link between sex and regionally specific brain connectivity, they do not support a clear-cut dimorphism in functional brain organization that is driven by sex alone.

Multimodal Parcellations and Extensive Behavioral Profiling Tackling the Hippocampus Gradient.

The hippocampus displays a complex organization and function that is perturbed in many neuropathologies. Histological work revealed a complex arrangement of subfields along the medial-lateral and the ventral-dorsal dimension, which contrasts with the anterior-posterior functional differentiation. The variety of maps has raised the need for an integrative multimodal view. We applied connectivity-based parcellation to 1) intrinsic connectivity 2) task-based connectivity, and 3) structural covariance, as complementary windows into structural and functional differentiation of the hippocampus. Strikingly, while functional properties (i.e., intrinsic and task-based) revealed similar partitions dominated by an anterior-posterior organization, structural covariance exhibited a hybrid pattern reflecting both functional and cytoarchitectonic subdivision. Capitalizing on the consistency of functional parcellations, we defined robust functional maps at different levels of partitions, which are openly available for the scientific community. Our functional maps demonstrated a head-body and tail partition, subdivided along the anterior-posterior and medial-lateral axis. Behavioral profiling of these fine partitions based on activation data indicated an emotion-cognition gradient along the anterior-posterior axis and additionally suggested a self-world-centric gradient supporting the role of the hippocampus in the construction of abstract representations for spatial navigation and episodic memory.

Evaluation of non-negative matrix factorization of grey matter in age prediction.

The relationship between grey matter volume (GMV) patterns and age can be captured by multivariate pattern analysis, allowing prediction of individuals' age based on structural imaging. Raw data, voxel-wise GMV and non-sparse factorization (with Principal Component Analysis, PCA) show good performance but do not promote relatively localized brain components for post-hoc examinations. Here we evaluated a non-negative matrix factorization (NNMF) approach to provide a reduced, but also interpretable representation of GMV data in age prediction frameworks in healthy and clinical populations. This examination was performed using three datasets: a multi-site cohort of life-span healthy adults, a single site cohort of older adults and clinical samples from the ADNI dataset with healthy subjects, participants with Mild Cognitive Impairment and patients with Alzheimer's disease (AD) subsamples. T1-weighted images were preprocessed with VBM8 standard settings to compute GMV values after normalization, segmentation and modulation for non-linear transformations only. Non-negative matrix factorization was computed on the GM voxel-wise values for a range of granularities (50-690 components) and LASSO (Least Absolute Shrinkage and Selection Operator) regression were used for age prediction. First, we compared the performance of our data compression procedure (i.e., NNMF) to various other approaches (i.e., uncompressed VBM data, PCA-based factorization and parcellation-based compression). We then investigated the impact of the granularity on the accuracy of age prediction, as well as the transferability of the factorization and model generalization across datasets. We finally validated our framework by examining age prediction in ADNI samples. Our results showed that our framework favorably compares with other approaches. They also demonstrated that the NNMF based factorization derived from one dataset could be efficiently applied to compress VBM data of another dataset and that granularities between 300 and 500 components give an optimal representation for age prediction. In addition to the good performance in healthy subjects our framework provided relatively localized brain regions as the features contributing to the prediction, thereby offering further insights into structural changes due to brain aging. Finally, our validation in clinical populations showed that our framework is sensitive to deviance from normal structural variations in pathological aging.

Prediction of outcome after aneurysmal subarachnoid haemorrhage using data from patient admission.

OBJECTIVES: The pathogenesis leading to poor functional outcome after aneurysmal subarachnoid haemorrhage (aSAH) is multifactorial and not fully understood. We evaluated a machine learning approach based on easily determinable clinical and CT perfusion (CTP) features in the course of patient admission to predict the functional outcome 6 months after ictus. METHODS: Out of 630 consecutive subarachnoid haemorrhage patients (2008-2015), 147 (mean age 54.3, 66.7% women) were retrospectively included (Inclusion: aSAH, admission within 24 h of ictus, CTP within 24 h of admission, documented modified Rankin scale (mRS) grades after 6 months. Exclusion: occlusive therapy before first CTP, previous aSAH, CTP not evaluable). A random forests model with conditional inference trees was optimised and trained on sex, age, World Federation of Neurosurgical Societies (WFNS) and modified Fisher grades, aneurysm in anterior vs. posterior circulation, early external ventricular drainage (EVD), as well as MTT and Tmax maximum, mean, standard deviation (SD), range, 75th quartile and interquartile range to predict dichotomised mRS (≤ 2; > 2). Performance was assessed using the balanced accuracy over the training and validation folds using 20 repeats of 10-fold cross-validation. RESULTS: In the final model, using 200 trees and the synthetic minority oversampling technique, median balanced accuracy was 84.4% (SD 0.7) over the training folds and 70.9% (SD 1.2) over the validation folds. The five most important features were the modified Fisher grade, age, MTT range, WFNS and early EVD. CONCLUSIONS: A random forests model trained on easily determinable features in the course of patient admission can predict the functional outcome 6 months after aSAH with considerable accuracy. KEY POINTS: • Features determinable in the course of admission of a patient with aneurysmal subarachnoid haemorrhage (aSAH) can predict the functional outcome 6 months after the occurrence of aSAH. • The top five predictive features were the modified Fisher grade, age, the mean transit time (MTT) range from computed tomography perfusion (CTP), the WFNS grade and the early necessity for an external ventricular drainage (EVD). • The range between the minimum and the maximum MTT may prove to be a valuable biomarker for detrimental functional outcome.

A simple plug-in bagging ensemble based on threshold-moving for classifying binary and multiclass imbalanced data.

Class imbalance presents a major hurdle in the application of classification methods. A commonly taken approach is to learn ensembles of classifiers using rebalanced data. Examples include bootstrap averaging (bagging) combined with either undersampling or oversampling of the minority class examples. However, rebalancing methods entail asymmetric changes to the examples of different classes, which in turn can introduce their own biases. Furthermore, these methods often require specifying the performance measure of interest a priori, i.e., before learning. An alternative is to employ the threshold moving technique, which applies a threshold to the continuous output of a model, offering the possibility to adapt to a performance measure a posteriori, i.e., a plug-in method. Surprisingly, little attention has been paid to this combination of a bagging ensemble and threshold-moving. In this paper, we study this combination and demonstrate its competitiveness. Contrary to the other resampling methods, we preserve the natural class distribution of the data resulting in well-calibrated posterior probabilities. Additionally, we extend the proposed method to handle multiclass data. We validated our method on binary and multiclass benchmark data sets by using both, decision trees and neural networks as base classifiers. We perform analyses that provide insights into the proposed method.

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.

Lifetime Exposure to Depression and Neuroimaging Measures of Brain Structure and Function.

Importance: Despite decades of neuroimaging studies reporting brain structural and functional alterations in depression, discrepancies in findings across studies and limited convergence across meta-analyses have raised questions about the consistency and robustness of the observed brain phenotypes. Objective: To investigate the associations between 6 operational criteria of lifetime exposure to depression and functional and structural neuroimaging measures. Design, Setting, and Participants: This cross-sectional study analyzed data from a UK Biobank cohort of individuals aged 45 to 80 years who were enrolled between January 1, 2014, and December 31, 2018. Participants included individuals with a lifetime exposure to depression and matched healthy controls without indications of psychosis, mental illness, behavior disorder, and disease of the nervous system. Six operational criteria of lifetime exposure to depression were evaluated: help seeking for depression; self-reported depression; antidepressant use; depression definition by Smith et al; hospital International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) diagnosis codes F32 and F33; and Composite International Diagnostic Interview Short Form score. Six increasingly restrictive depression definitions and groups were defined based on the 6 depression criteria, ranging from meeting only 1 criterion to meeting all 6 criteria. Data were analyzed between January and October 2022. Main Outcomes and Measures: Functional measures were calculated using voxel-wise fractional amplitude of low-frequency fluctuation (fALFF), global correlation (GCOR), and local correlation (LCOR). Structural measures were calculated using gray matter volume (GMV). Results: The study included 20 484 individuals with lifetime depression (12 645 females [61.7%]; mean [SD] age, 63.91 [7.60] years) and 25 462 healthy controls (14 078 males [55.3%]; mean [SD] age, 65.05 [7.8] years). Across all depression criteria, individuals with lifetime depression displayed regionally consistent decreases in fALFF, LCOR, and GCOR (Cohen d range, -0.53 [95% CI, -0.88 to -0.15] to -0.04 [95% CI, -0.07 to -0.01]) but not in GMV (Cohen d range, -0.47 [95 % CI, -0.75 to -0.12] to 0.26 [95% CI, 0.15-0.37]). Hospital ICD-10 diagnosis codes F32 and F33 (median [IQR] difference in effect sizes, -0.14 [-0.17 to -0.11]) and antidepressant use (median [IQR] difference in effect sizes, -0.12 [-0.16 to -0.10]) were criteria associated with the most pronounced alterations. Conclusions and Relevance: Results of this cross-sectional study indicate that lifetime exposure to depression was associated with robust functional changes, with a more restrictive depression definition revealing more pronounced alterations. Different inclusion criteria for depression may be associated with the substantial variation in imaging findings reported in the literature.