What predicts citation counts and translational impact in headache research? A machine learning analysis.

BACKGROUND: We aimed to develop the first machine learning models to predict citation counts and the translational impact, defined as inclusion in guidelines or policy documents, of headache research, and assess which factors are most predictive. METHODS: Bibliometric data and the titles, abstracts, and keywords from 8600 publications in three headache-oriented journals from their inception to 31 December 2017 were used. A series of machine learning models were implemented to predict three classes of 5-year citation count intervals (0-5, 6-14 and, >14 citations); and the translational impact of a publication. Models were evaluated out-of-sample with area under the receiver operating characteristics curve (AUC). RESULTS: The top performing gradient boosting model predicted correct citation count class with an out-of-sample AUC of 0.81. Bibliometric data such as page count, number of references, first and last author citation counts and h-index were among the most important predictors. Prediction of translational impact worked optimally when including both bibliometric data and information from the title, abstract and keywords, reaching an out-of-sample AUC of 0.71 for the top performing random forest model. CONCLUSION: Citation counts are best predicted by bibliometric data, while models incorporating both bibliometric data and publication content identifies the translational impact of headache research.

Computational limits to the legibility of the imaged human brain.

Our knowledge of the organisation of the human brain at the population-level is yet to translate into power to predict functional differences at the individual-level, limiting clinical applications and casting doubt on the generalisability of inferred mechanisms. It remains unknown whether the difficulty arises from the absence of individuating biological patterns within the brain, or from limited power to access them with the models and compute at our disposal. Here we comprehensively investigate the resolvability of such patterns with data and compute at unprecedented scale. Across 23 810 unique participants from UK Biobank, we systematically evaluate the predictability of 25 individual biological characteristics, from all available combinations of structural and functional neuroimaging data. Over 4526 GPU*hours of computation, we train, optimize, and evaluate out-of-sample 700 individual predictive models, including fully-connected feed-forward neural networks of demographic, psychological, serological, chronic disease, and functional connectivity characteristics, and both uni- and multi-modal 3D convolutional neural network models of macro- and micro-structural brain imaging. We find a marked discrepancy between the high predictability of sex (balanced accuracy 99.7%), age (mean absolute error 2.048 years, R2 0.859), and weight (mean absolute error 2.609Kg, R2 0.625), for which we set new state-of-the-art performance, and the surprisingly low predictability of other characteristics. Neither structural nor functional imaging predicted an individual's psychology better than the coincidence of common chronic disease (p < 0.05). Serology predicted chronic disease (p < 0.05) and was best predicted by it (p < 0.001), followed by structural neuroimaging (p < 0.05). Our findings suggest either more informative imaging or more powerful models will be needed to decipher individual level characteristics from the human brain. We make our models and code openly available.

Voxel-based dysconnectomic brain morphometry with computed tomography in Down syndrome.

OBJECTIVE: Alzheimer's disease (AD) is a major health concern for aging adults with Down syndrome (DS), but conventional diagnostic techniques are less reliable in those with severe baseline disability. Likewise, acquisition of magnetic resonance imaging to evaluate cerebral atrophy is not straightforward, as prolonged scanning times are less tolerated in this population. Computed tomography (CT) scans can be obtained faster, but poor contrast resolution limits its function for morphometric analysis. We implemented an automated analysis of CT scans to characterize differences across dementia stages in a cross-sectional study of an adult DS cohort. METHODS: CT scans of 98 individuals were analyzed using an automatic algorithm. Voxel-based correlations with clinical dementia stages and AD plasma biomarkers (phosphorylated tau-181 and neurofilament light chain) were identified, and their dysconnectomic patterns delineated. RESULTS: Dementia severity was negatively correlated with gray (GM) and white matter (WM) volumes in temporal lobe regions, including parahippocampal gyri. Dysconnectome analysis revealed an association between WM loss and temporal lobe GM volume reduction. AD biomarkers were negatively associated with GM volume in hippocampal and cingulate gyri. INTERPRETATION: Our automated algorithm and novel dysconnectomic analysis of CT scans successfully described brain morphometric differences related to AD in adults with DS, providing a new avenue for neuroimaging analysis in populations for whom magnetic resonance imaging is difficult to obtain.

Characterizing phonemic fluency by transfer learning with deep language models.

Though phonemic fluency tasks are traditionally indexed by the number of correct responses, the underlying disorder may shape the specific choice of words-both correct and erroneous. We report the first comprehensive qualitative analysis of incorrect and correct words generated on the phonemic ('S') fluency test, in a large sample of patients (n = 239) with focal, unilateral frontal or posterior lesions and healthy controls (n = 136). We conducted detailed qualitative analyses of the single words generated in the phonemic fluency task using categorical descriptions for different types of errors, low-frequency words and clustering/switching. We further analysed patients' and healthy controls' entire sequences of words by employing stochastic block modelling of Generative Pretrained Transformer 3-based deep language representations. We conducted predictive modelling to investigate whether deep language representations of word sequences improved the accuracy of detecting the presence of frontal lesions using the phonemic fluency test. Our qualitative analyses of the single words generated revealed several novel findings. For the different types of errors analysed, we found a non-lateralized frontal effect for profanities, left frontal effects for proper nouns and permutations and a left posterior effect for perseverations. For correct words, we found a left frontal effect for low-frequency words. Our novel large language model-based approach found five distinct communities whose varied word selection patterns reflected characteristic demographic and clinical features. Predictive modelling showed that a model based on Generative Pretrained Transformer 3-derived word sequence representations predicted the presence of frontal lesions with greater fidelity than models of native features. Our study reveals a characteristic pattern of phonemic fluency responses produced by patients with frontal lesions. These findings demonstrate the significant inferential and diagnostic value of characterizing qualitative features of phonemic fluency performance with large language models and stochastic block modelling.

Federated learning for medical imaging radiology.

Federated learning (FL) is gaining wide acceptance across the medical AI domains. FL promises to provide a fairly acceptable clinical-grade accuracy, privacy, and generalisability of machine learning models across multiple institutions. However, the research on FL for medical imaging AI is still in its early stages. This paper presents a review of recent research to outline the difference between state-of-the-art [SOTA] (published literature) and state-of-the-practice [SOTP] (applied research in realistic clinical environments). Furthermore, the review outlines the future research directions considering various factors such as data, learning models, system design, governance, and human-in-loop to translate the SOTA into SOTP and effectively collaborate across multiple institutions.

Examining patient benefit.

Healthcare policy, clinical practice and clinical research all declare patient benefit as their avowed aim. Yet, the conceptual question of what exactly constitutes patient benefit has received much less attention than the practical means of realising it. Currently, three key areas of conceptual unclarity make the achieved, real-world impact hard to quantify and disconnect it from the magnitude of the practical endeavour: (1) the distinction between objective and subjective benefit, (2) the relation between individual and population measures of benefit, and (3) the optimal measurement of benefit in research studies. A philosophical understanding of wellbeing is required to clarify these problems. Adopting a rigorous philosophical framework makes apparent that the differing goals of clinicians, researchers and research funders may make differing conceptions of patient benefit appropriate. A framework is proposed for developing rigour in methods for specifying and measuring patient benefit, and for matching benefit measures to different contexts.

Latent Transformer Models for out-of-distribution detection.

Any clinically-deployed image-processing pipeline must be robust to the full range of inputs it may be presented with. One popular approach to this challenge is to develop predictive models that can provide a measure of their uncertainty. Another approach is to use generative modelling to quantify the likelihood of inputs. Inputs with a low enough likelihood are deemed to be out-of-distribution and are not presented to the downstream predictive model. In this work, we evaluate several approaches to segmentation with uncertainty for the task of segmenting bleeds in 3D CT of the head. We show that these models can fail catastrophically when operating in the far out-of-distribution domain, often providing predictions that are both highly confident and wrong. We propose to instead perform out-of-distribution detection using the Latent Transformer Model: a VQ-GAN is used to provide a highly compressed latent representation of the input volume, and a transformer is then used to estimate the likelihood of this compressed representation of the input. We demonstrate this approach can identify images that are both far- and near- out-of-distribution, as well as provide spatial maps that highlight the regions considered to be out-of-distribution. Furthermore, we find a strong relationship between an image's likelihood and the quality of a model's segmentation on it, demonstrating that this approach is viable for filtering out unsuitable images.

Brain tumour genetic network signatures of survival.

Tumour heterogeneity is increasingly recognized as a major obstacle to therapeutic success across neuro-oncology. Gliomas are characterized by distinct combinations of genetic and epigenetic alterations, resulting in complex interactions across multiple molecular pathways. Predicting disease evolution and prescribing individually optimal treatment requires statistical models complex enough to capture the intricate (epi)genetic structure underpinning oncogenesis. Here, we formalize this task as the inference of distinct patterns of connectivity within hierarchical latent representations of genetic networks. Evaluating multi-institutional clinical, genetic and outcome data from 4023 glioma patients over 14 years, across 12 countries, we employ Bayesian generative stochastic block modelling to reveal a hierarchical network structure of tumour genetics spanning molecularly confirmed glioblastoma, IDH-wildtype; oligodendroglioma, IDH-mutant and 1p/19q codeleted; and astrocytoma, IDH-mutant. Our findings illuminate the complex dependence between features across the genetic landscape of brain tumours and show that generative network models reveal distinct signatures of survival with better prognostic fidelity than current gold standard diagnostic categories.

Phenotype of new daily persistent headache: subtypes and comparison to transformed chronic daily headache.

BACKGROUND: It is unknown whether new daily persistent headache (NDPH) is a single disorder or heterogenous group of disorders, and whether it is a unique disorder from chronic migraine and chronic tension-type headache. We describe a large group of patients with primary NDPH, compare its phenotype to transformed chronic daily headache (T-CDH), and use cluster analysis to reveal potential sub-phenotypes in the NDPH group. METHODS: We performed a case-control study using prospectively collected clinical data in patients with primary NDPH and T-CDH (encompassing chronic migraine and chronic tension-type headache). We used logistic regression with propensity score matching to compare demographics, phenotype, comorbidities, and treatment responses between NDPH and T-CDH. We used K-means cluster analysis with Gower distance to identify sub-clusters in the NDPH group based on a combination of demographics, phenotype, and comorbidities. RESULTS: We identified 366 patients with NDPH and 696 with T-CDH who met inclusion criteria. Patients with NDPH were less likely to be female (62.6% vs. 73.3%, p < 0.001). Nausea, vomiting, photophobia, phonophobia, motion sensitivity, vertigo, and cranial autonomic symptoms were all significantly less frequent in NDPH than T-CDH (p value for all < 0.001). Acute treatments appeared less effective in NDPH than T-CDH, and medication overuse was less common (16% vs. 42%, p < 0.001). Response to most classes of oral preventive treatments was poor in both groups. The most effective treatment in NDPH was doselupin in 45.7% patients (95% CI 34.8-56.5%). Cluster analysis identified three subgroups of NDPH. Cluster 1 was older, had a high proportion of male patients, and less severe headaches. Cluster 2 was predominantly female, had severe headaches, and few associated symptoms. Cluster 3 was predominantly female with a high prevalence of migrainous symptoms and headache triggers. CONCLUSIONS: Whilst there is overlap in the phenotype of NDPH and T-CDH, the differences in migrainous, cranial autonomic symptoms, and vulnerability to medication overuse suggest that they are not the same disorder. NDPH may be fractionated into three sub-phenotypes, which require further investigation.

The human cost of ethical artificial intelligence.

Foundational models such as ChatGPT critically depend on vast data scales the internet uniquely enables. This implies exposure to material varying widely in logical sense, factual fidelity, moral value, and even legal status. Whereas data scaling is a technical challenge, soluble with greater computational resource, complex semantic filtering cannot be performed reliably without human intervention: the self-supervision that makes foundational models possible at least in part presupposes the abilities they seek to acquire. This unavoidably introduces the need for large-scale human supervision-not just of training input but also model output-and imbues any model with subjectivity reflecting the beliefs of its creator. The pressure to minimize the cost of the former is in direct conflict with the pressure to maximise the quality of the latter. Moreover, it is unclear how complex semantics, especially in the realm of the moral, could ever be reduced to an objective function any machine could plausibly maximise. We suggest the development of foundational models necessitates urgent innovation in quantitative ethics and outline possible avenues for its realisation.

Brain tumour segmentation with incomplete imaging data.

Progress in neuro-oncology is increasingly recognized to be obstructed by the marked heterogeneity-genetic, pathological, and clinical-of brain tumours. If the treatment susceptibilities and outcomes of individual patients differ widely, determined by the interactions of many multimodal characteristics, then large-scale, fully-inclusive, richly phenotyped data-including imaging-will be needed to predict them at the individual level. Such data can realistically be acquired only in the routine clinical stream, where its quality is inevitably degraded by the constraints of real-world clinical care. Although contemporary machine learning could theoretically provide a solution to this task, especially in the domain of imaging, its ability to cope with realistic, incomplete, low-quality data is yet to be determined. In the largest and most comprehensive study of its kind, applying state-of-the-art brain tumour segmentation models to large scale, multi-site MRI data of 1251 individuals, here we quantify the comparative fidelity of automated segmentation models drawn from MR data replicating the various levels of completeness observed in real life. We demonstrate that models trained on incomplete data can segment lesions very well, often equivalently to those trained on the full completement of images, exhibiting Dice coefficients of 0.907 (single sequence) to 0.945 (complete set) for whole tumours and 0.701 (single sequence) to 0.891 (complete set) for component tissue types. This finding opens the door both to the application of segmentation models to large-scale historical data, for the purpose of building treatment and outcome predictive models, and their application to real-world clinical care. We further ascertain that segmentation models can accurately detect enhancing tumour in the absence of contrast-enhancing imaging, quantifying the burden of enhancing tumour with an R 2 > 0.97, varying negligibly with lesion morphology. Such models can quantify enhancing tumour without the administration of intravenous contrast, inviting a revision of the notion of tumour enhancement if the same information can be extracted without contrast-enhanced imaging. Our analysis includes validation on a heterogeneous, real-world 50 patient sample of brain tumour imaging acquired over the last 15 years at our tertiary centre, demonstrating maintained accuracy even on non-isotropic MRI acquisitions, or even on complex post-operative imaging with tumour recurrence. This work substantially extends the translational opportunity for quantitative analysis to clinical situations where the full complement of sequences is not available and potentially enables the characterization of contrast-enhanced regions where contrast administration is infeasible or undesirable.

Prediction of Stroke Onset Time With Combined Fast High-Resolution Magnetic Resonance Spectroscopic and Quantitative T2 Mapping.

OBJECTIVE: The purpose of this work is to develop a multispectral imaging approach that combines fast high-resolution 3D magnetic resonance spectroscopic imaging (MRSI) and fast quantitative T2 mapping to capture the multifactorial biochemical changes within stroke lesions and evaluate its potentials for stroke onset time prediction. METHODS: Special imaging sequences combining fast trajectories and sparse sampling were used to obtain whole-brain maps of both neurometabolites (2.0 × 3.0 × 3.0 mm3) and quantitative T2 values (1.9 × 1.9 × 3.0 mm3) within a 9-minute scan. Participants with ischemic stroke at hyperacute (0-24 h, n = 23) or acute (24 h-7d, n = 33) phase were recruited in this study. Lesion N-acetylaspartate (NAA), lactate, choline, creatine, and T2 signals were compared between groups and correlated with patient symptomatic duration. Bayesian regression analyses were employed to compare the predictive models of symptomatic duration using multispectral signals. RESULTS: In both groups, increased T2 and lactate levels, as well as decreased NAA and choline levels were detected within the lesion (all p < 0.001). Changes in T2, NAA, choline, and creatine signals were correlated with symptomatic duration for all patients (all p < 0.005). Predictive models of stroke onset time combining signals from MRSI and T2 mapping achieved the best performance (hyperacute: R2 = 0.438; all: R2 = 0.548). CONCLUSION: The proposed multispectral imaging approach provides a combination of biomarkers that index early pathological changes after stroke in a clinical-feasible time and improves the assessment of the duration of cerebral infarction. SIGNIFICANCE: Developing accurate and efficient neuroimaging techniques to provide sensitive biomarkers for prediction of stroke onset time is of great importance for maximizing the proportion of patients eligible for therapeutic intervention. The proposed method provides a clinically feasible tool for the assessment of symptom onset time post ischemic stroke, which will help guide time-sensitive clinical management.

Forecasting migraine with machine learning based on mobile phone diary and wearable data.

INTRODUCTION: Triggers, premonitory symptoms and physiological changes occur in the preictal migraine phase and may be used in models for forecasting attacks. Machine learning is a promising option for such predictive analytics. The objective of this study was to explore the utility of machine learning to forecast migraine attacks based on preictal headache diary entries and simple physiological measurements. METHODS: In a prospective development and usability study 18 patients with migraine completed 388 headache diary entries and self-administered app-based biofeedback sessions wirelessly measuring heart rate, peripheral skin temperature and muscle tension. Several standard machine learning architectures were constructed to forecast headache the subsequent day. Models were scored with area under the receiver operating characteristics curve. RESULTS: Two-hundred-and-ninety-five days were included in the predictive modelling. The top performing model, based on random forest classification, achieved an area under the receiver operating characteristics curve of 0.62 in a hold-out partition of the dataset. DISCUSSION: In this study we demonstrate the utility of using mobile health apps and wearables combined with machine learning to forecast headache. We argue that high-dimensional modelling may greatly improve forecasting and discuss important considerations for future design of forecasting models using machine learning and mobile health data.

A framework for focal and connectomic mapping of transiently disrupted brain function.

The distributed nature of the neural substrate, and the difficulty of establishing necessity from correlative data, combine to render the mapping of brain function a far harder task than it seems. Methods capable of combining connective anatomical information with focal disruption of function are needed to disambiguate local from global neural dependence, and critical from merely coincidental activity. Here we present a comprehensive framework for focal and connective spatial inference based on sparse disruptive data, and demonstrate its application in the context of transient direct electrical stimulation of the human medial frontal wall during the pre-surgical evaluation of patients with focal epilepsy. Our framework formalizes voxel-wise mass-univariate inference on sparsely sampled data within the statistical parametric mapping framework, encompassing the analysis of distributed maps defined by any criterion of connectivity. Applied to the medial frontal wall, this transient dysconnectome approach reveals marked discrepancies between local and distributed associations of major categories of motor and sensory behaviour, revealing differentiation by remote connectivity to which purely local analysis is blind. Our framework enables disruptive mapping of the human brain based on sparsely sampled data with minimal spatial assumptions, good statistical efficiency, flexible model formulation, and explicit comparison of local and distributed effects.

Latent disconnectome prediction of long-term cognitive-behavioural symptoms in stroke.

Stroke significantly impacts the quality of life. However, the long-term cognitive evolution in stroke is poorly predictable at the individual level. There is an urgent need to better predict long-term symptoms based on acute clinical neuroimaging data. Previous works have demonstrated a strong relationship between the location of white matter disconnections and clinical symptoms. However, rendering the entire space of possible disconnection-deficit associations optimally surveyable will allow for a systematic association between brain disconnections and cognitive-behavioural measures at the individual level. Here we present the most comprehensive framework, a composite morphospace of white matter disconnections (disconnectome) to predict neuropsychological scores 1 year after stroke. Linking the latent disconnectome morphospace to neuropsychological outcomes yields biological insights that are available as the first comprehensive atlas of disconnectome-deficit relations across 86 scores-a Neuropsychological White Matter Atlas. Our novel predictive framework, the Disconnectome Symptoms Discoverer, achieved better predictivity performances than six other models, including functional disconnection, lesion topology and volume modelling. Out-of-sample prediction derived from this atlas presented a mean absolute error below 20% and allowed personalize neuropsychological predictions. Prediction on an external cohort achieved an R2 = 0.201 for semantic fluency. In addition, training and testing were replicated on two external cohorts achieving an R2 = 0.18 for visuospatial performance. This framework is available as an interactive web application (http://disconnectomestudio.bcblab.com) to provide the foundations for a new and practical approach to modelling cognition in stroke. We hope our atlas and web application will help to reduce the burden of cognitive deficits on patients, their families and wider society while also helping to tailor future personalized treatment programmes and discover new targets for treatments. We expect our framework's range of assessments and predictive power to increase even further through future crowdsourcing.

Predicting the Onset of Ischemic Stroke With Fast High-Resolution 3D MR Spectroscopic Imaging.

BACKGROUND: Neurometabolite concentrations provide a direct index of infarction progression in stroke. However, their relationship with stroke onset time remains unclear. PURPOSE: To assess the temporal dynamics of N-acetylaspartate (NAA), creatine, choline, and lactate and estimate their value in predicting early (<6 hours) vs. late (6-24 hours) hyperacute stroke groups. STUDY TYPE: Cross-sectional cohort. POPULATION: A total of 73 ischemic stroke patients scanned at 1.8-302.5 hours after symptom onset, including 25 patients with follow-up scans. FIELD STRENGTH/SEQUENCE: A 3 T/magnetization-prepared rapid acquisition gradient echo sequence for anatomical imaging, diffusion-weighted imaging and fluid-attenuated inversion recovery imaging for lesion delineation, and 3D MR spectroscopic imaging (MRSI) for neurometabolic mapping. ASSESSMENT: Patients were divided into hyperacute (0-24 hours), acute (24 hours to 1 week), and subacute (1-2 weeks) groups, and into early (<6 hours) and late (6-24 hours) hyperacute groups. Bayesian logistic regression was used to compare classification performance between early and late hyperacute groups by using different combinations of neurometabolites as inputs. STATISTICAL TESTS: Linear mixed effects modeling was applied for group-wise comparisons between NAA, creatine, choline, and lactate. Pearson's correlation analysis was used for neurometabolites vs. time. P < 0.05 was considered statistically significant. RESULTS: Lesional NAA and creatine were significantly lower in subacute than in acute stroke. The main effects of time were shown on NAA (F = 14.321) and creatine (F = 12.261). NAA was significantly lower in late than early hyperacute patients, and was inversely related to time from symptom onset across both groups (r = -0.440). The decrease of NAA and increase of lactate were correlated with lesion volume (NAA: r = -0.472; lactate: r = 0.366) in hyperacute stroke. Discrimination was improved by combining NAA, creatine, and choline signals (area under the curve [AUC] = 0.90). DATA CONCLUSION: High-resolution 3D MRSI effectively assessed the neurometabolite changes and discriminated early and late hyperacute stroke lesions. EVIDENCE LEVEL: 1. TECHNICAL EFFICACY: Stage 2.

Predicting mortality in acutely hospitalised older patients: the impact of model dimensionality.

BACKGROUND: The prediction of long-term mortality following acute illness can be unreliable for older patients, inhibiting the delivery of targeted clinical interventions. The difficulty plausibly arises from the complex, multifactorial nature of the underlying biology in this population, which flexible, multimodal models based on machine learning may overcome. Here, we test this hypothesis by quantifying the comparative predictive fidelity of such models in a large consecutive sample of older patients acutely admitted to hospital and characterise their biological support. METHODS: A set of 804 admission episodes involving 616 unique patients with a mean age of 84.5 years consecutively admitted to the Acute Geriatric service at University College Hospital were identified, in whom clinical diagnoses, blood tests, cognitive status, computed tomography of the head, and mortality within 600 days after admission were available. We trained and evaluated out-of-sample an array of extreme gradient boosted trees-based predictive models of incrementally greater numbers of investigational modalities and modelled features. Both linear and non-linear associations with investigational features were quantified. RESULTS: Predictive models of mortality showed progressively increasing fidelity with greater numbers of modelled modalities and dimensions. The area under the receiver operating characteristic curve rose from 0.67 (sd = 0.078) for age and sex to 0.874 (sd = 0.046) for the most comprehensive model. Extracranial bone and soft tissue features contributed more than intracranial features towards long-term mortality prediction. The anterior cingulate and angular gyri, and serum albumin, were the greatest intracranial and biochemical model contributors respectively. CONCLUSIONS: High-dimensional, multimodal predictive models of mortality based on routine clinical data offer higher predictive fidelity than simpler models, facilitating individual level prognostication and interventional targeting. The joint contributions of both extracranial and intracranial features highlight the potential importance of optimising somatic as well as neural functions in healthy ageing. Our findings suggest a promising path towards a high-fidelity, multimodal index of frailty.

Orienting to fear under transient focal disruption of the human amygdala.

Responding to threat is under strong survival pressure, promoting the evolution of systems highly optimized for the task. Though the amygdala is implicated in 'detecting' threat, its role in the action that immediately follows-'orienting'-remains unclear. Critical to mounting a targeted response, such early action requires speed, accuracy, and resilience optimally achieved through conserved, parsimonious, dedicated systems, insured against neural loss by a parallelized functional organization. These characteristics tend to conceal the underlying substrate not only from correlative methods but also from focal disruption over time scales long enough for compensatory adaptation to take place. In a study of six patients with intracranial electrodes temporarily implanted for the clinical evaluation of focal epilepsy, we investigated gaze orienting to fear during focal, transient, unilateral direct electrical disruption of the amygdala. We showed that the amygdala is necessary for rapid gaze shifts towards faces presented in the contralateral hemifield regardless of their emotional expression, establishing its functional lateralization. Behaviourally dissociating the location of presented fear from the direction of the response, we implicated the amygdala not only in detecting contralateral faces, but also in automatically orienting specifically towards fearful ones. This salience-specific role was demonstrated within a drift-diffusion model of action to manifest as an orientation bias towards the location of potential threat. Pixel-wise analysis of target facial morphology revealed scleral exposure as its primary driver, and induced gamma oscillations-obtained from intracranial local field potentials-as its time-locked electrophysiological correlate. The amygdala is here reconceptualized as a functionally lateralized instrument of early action, reconciling previous conflicting accounts confined to detection, and revealing a neural organisation analogous to the superior colliculus, with which it is phylogenetically kin. Greater clarity on its role has the potential to guide therapeutic resection, still frequently complicated by impairments of cognition and behaviour related to threat, and inform novel focal stimulation techniques for the management of neuropsychiatric conditions.

Equitable modelling of brain imaging by counterfactual augmentation with morphologically constrained 3D deep generative models.

We describe CounterSynth, a conditional generative model of diffeomorphic deformations that induce label-driven, biologically plausible changes in volumetric brain images. The model is intended to synthesise counterfactual training data augmentations for downstream discriminative modelling tasks where fidelity is limited by data imbalance, distributional instability, confounding, or underspecification, and exhibits inequitable performance across distinct subpopulations. Focusing on demographic attributes, we evaluate the quality of synthesised counterfactuals with voxel-based morphometry, classification and regression of the conditioning attributes, and the Fréchet inception distance. Examining downstream discriminative performance in the context of engineered demographic imbalance and confounding, we use UK Biobank and OASIS magnetic resonance imaging data to benchmark CounterSynth augmentation against current solutions to these problems. We achieve state-of-the-art improvements, both in overall fidelity and equity. The source code for CounterSynth is available at https://github.com/guilherme-pombo/CounterSynth.