Aperture Neuro

MICAFlow is a fully automated MRI preprocessing pipeline designed to translate advanced neuroimaging workflows from research into routine clinical practice. The pipeline emphasizes speed, robustness, and ease of use, focusing on structural and diffusion MRI. Key innovations include a Label-Augmented Modality-Agnostic Registration (LAMAReg) technique driven by deep learning segmentations for reliable cross-modal alignment, integration of state-of-the-art distortion corrections, and adherence to reproducible standards (Snakemake workflow, BIDSApp specifications). We describe the design of MICAFlow and evaluate its performance across heterogeneous datasets. First, accessibility: MICAFlow processes a multimodal MRI exam in minutes with clinically accessible hardware and without requiring GPU access, making it feasible for same-day clinical use. Second, registration accuracy: LAMAReg achieves cutting-edge multi-modal registration accuracy, yielding accurate alignment of diffusion MRI, FLAIR, and intra-subject T1-weighted images while remaining generally robust to common artifacts. Third, data reliability: Using identifiability, we show MICAFlow maintains consistent performance across diverse datasets, including subjects with pathology, and is closely comparable to contemporary pipelines. In sum, MICAFlows combination of machine learning and efficient workflows produces research-grade data quality with clinical-grade speed. This work demonstrates that advanced MRI preprocessing can be done fast and robustly, helping close the gap between research neuroimaging and broad clinical application of quantitative MRI techniques. The source code for MICAFlow is available here: https://github.com/MICA-MNI/micaflow, and for LAMAReg here: https://github.com/MICA-MNI/LAMAReg.

Neuroimaging data management and processing are tedious and error-prone, prompting reproducibility concerns. Globally, studies with heterogeneous infrastructure and governance policies lead to eclectic data processing and sharing, necessitating standardization of data workflows to ensure reusability and comparability of multi-centric datasets. The Nipoppy neuroinformatics framework facilitates such standardization by combining specification, protocol, and software to manage study-level data workflows. With its adoption, researchers can share standardized, derived datasets enabling efficient, reproducible, and inclusive research.

BackgroundPooling multi-site MRI data is essential for well-powered neuroimaging analyses, particularly in Huntingtons disease (HD), where large cohorts are needed to study disease-stage heterogeneity and subtle progressive neuroanatomical change. However, scanner-related variability hinders direct data pooling, confounding image-level methods such as voxel-based morphometry (VBM). Superpixel-ComBat (SP-ComBat), a voxel-level image-harmonization framework, effectively removes scanner effects but depends on traveling-subject data that are rarely available retrospectively. We extend SP-ComBat to unpaired multi-site T1-weighted MRI by introducing a pseudo-pairing framework that leverages demographically matched controls across scanners as surrogate traveling subjects. MethodsTwo pipelines were developed to estimate scanner effects under retrospective constraints: pipeline 1 used a small set of well-matched pseudo-pairs (n = 4) with bootstrap resampling to address scanners with limited sample sizes, while pipeline 2 used the recommended number of pseudo-pairs (n = 16) without resampling. Pseudo-pair images were parcellated into 3D-superpixels, and ComBat was applied within clusters to estimate scanner-specific adjustments for native-space harmonization. Pipeline performance was assessed in a representative multi-study dataset comprising six scanners from three HD cohorts (HD-YAS, HD-CSF, TRACK-HD; N = 144) and replicated in the full multi-study dataset (FMD; N = 548). ResultsBoth pipelines improved image quality, aligned scanner-specific intensities, and preserved disease-related structural patterns. Pipeline 2 showed superior parameter re-estimation stability and was selected for the FMD. Harmonization eliminated systematic segmentation errors, enabled a single unified VBM pipeline across scanners, and increased sensitivity to HD-related voxel-wise atrophy. ConclusionsSP-ComBat was effectively adapted for harmonization of unpaired multi-site structural MRI, reducing scanner bias while preserving biological variability and supporting unified VBM analyses across scanners.

Understanding vascular contributions to disease is a major unmet need. White matter lesions (WML) are an accepted imaging marker of cerebral small vessel disease, giving insights into its related pathologies. A unified approach for WML analyses in large multi-site data is lacking despite the need for pooling of data to overcome the limitations of often small heterogenous MRI studies which make subtyping and identifying patterns within disease groups difficult. Our ENIGMA-PD-WML pipeline is an open-source containerized pipeline containing all the code and packages required for pre-processing, processing and post-processing of T1-weighted and FLAIR data, outputting accurate and reproducible binary WML maps using a UNet approach. The pipeline provides a standardized image analysis approach for WML and outputs data in both native and MNI space to allow for sharing and pooling of data from multiple sites for large-data analysis. In addition to a reliable standardized approach for WML segmentation, key priorities when developing the pipeline included: usability, i.e., requiring minimal manual input and technical expertise to use, and suitability to run on various MRI scanners and acquisition parameters as is common in multi-site data. This paper describes the pipeline in detail, with rationale for each step, providing transparency and facilitating its usage to overcome reproducibility issues in large-scale WML analyses.

Machine-learning models are increasingly used to predict cognitive and clinical outcomes from neuroimaging data, yet challenges in fairness and generalizability remain. Large-scale datasets are often racially and ethnically imbalanced, leading to systematic performance disparities, with models typically achieving higher accuracy for majority populations represented in the training data. In this study, we evaluated whether supervised domain adaptation methods--including balanced weighting, two-stage TrAdaBoost, feature augmentation with SrcOnly prediction, and linear interpolation--can mitigate these biases. Using the ABCD dataset, we assessed whether models trained on 80 MRI measures from White American participants could generalize more effectively to African American participants. All domain adaptation methods reduced prediction error for African American participants, particularly for MRI modalities with large baseline disparities (e.g., structural MRI), while offering limited improvements where initial gaps were smaller (e.g., functional connectivity). Among the approaches, balanced weighting performed best and remained stable and beneficial even when only 10 African American participants were used to adapt the original model trained exclusively on White American participants. These findings suggest that simple, low-cost strategies can effectively reduce cross-ethnic performance gaps and improve equity in predictive neuroimaging, offering a practical path forward for future neuroimaging predictive biomarkers. Significant StatementLarge-scale neuroimaging datasets increasingly enable machine-learning models to predict cognitive and clinical outcomes; however, these datasets are often ethnically/racially imbalanced. As a result, predictive models tend to generalize poorly to underrepresented populations. We demonstrate that, across 80 MRI phenotypes, a class of machine-learning approaches collectively known as supervised domain adaptation can substantially reduce cross-ethnicity disparities in neuroimaging-based cognitive prediction, even when only limited data from underrepresented groups are available. Among the methods evaluated, balanced weighting achieved the best performance while maintaining low computational cost. Together, these findings provide a practical and scalable framework for improving fairness and generalizability in neuroimaging-based machine learning under realistic conditions of ethnic/racial imbalance.

IntroductionPostmortem human brain magnetic resonance imaging (MRI) offers a unique opportunity to study finer neuroanatomical details and enables direct correlations with gold standard histological and immunohistochemical assessments. However, to prevent tissue decay, postmortem brains are preserved in fixative solutions which can alter tissue properties and exert substantial impacts on the MRI signals. The present study investigates the impact of formalin fixation, the most commonly used solution for postmortem human brain preservation, on different quantitative MRI contrasts. Methods142 intact human brain hemispheres immersed in 10% formalin for a range of fixation durations (between 0 days and 20 years) were imaged in a 3T MRI scanner. A subset of 10 brains were further scanned repeatedly at days 0, 3, 10, 20, 30, 60, 90, and 120 to allow for better characterization of the initial transient effects of fixation. Voxel-wise T1 and T2* relaxation, T1/T2 ratio, and myelin water fraction (MWF) maps were generated for each specimen and timepoint, and linear and nonlinear models were used to examine the spatiotemporal changes associated with progressive fixation. ResultsAll investigated metrics were significantly impacted by formalin fixation, albeit at different rates and with differing regional patterns. T1 and T2* relaxation time decreased as a result of progressive fixation, whereas T1/T2 ratio and MWF measures increased. T1 relaxation and T1/T2 ratio showed nonlinear patterns with initially accelerated changes that decelerate in the first few months, whereas T2* relaxation and MWF changes followed a more linear trend. ConclusionFormaldehyde fixation exerts systematic changes on quantitative MRI signals that can be modeled and adjusted for to allow for harmonized comparisons of MRI metrics across brains fixed for differing durations. The distinct temporal trajectories observed across metrics highlight the need to account for fixation duration in study design and downstream analyses, particularly when integrating datasets acquired under heterogeneous conditions. Our findings provide a quantitative framework for correcting fixation-induced biases, thereby improving the interpretability and reproducibility of postmortem MRI studies.

Compared to traditional gross volumetrics, surface- based models provide greater spatial precision for understanding brain alterations related to developmental, neurological, and psychiatric disorders. Large-scale brain initiatives are combining data from around the world to discover and improve illness- related brain markers. Here, we present a toolkit for 3D brain geometry analysis aimed at addressing key challenges facing large- scale neuroimaging studies. Our framework incorporates scalable methods for multisite data integration, site-specific confound correction, accelerated statistical modeling, interpretable machine learning, and interactive results visualization. The toolkit was tested on data from 21 independently collected study samples participating in the ENIGMA Bipolar Disorder Working Group (N = 3,373). Compared to traditional volume features, we show how subcortical shape measures can be combined across study sites to capture spatially complex differences between diagnostic groups and associations with common treatments. Statistical modeling was accelerated using the Fast and Efficient Mixed- Effects Algorithm (FEMA) and achieved a 16-fold reduction in computation time compared to traditional approaches. Machine learning models showed shape features may provide greater predictive performance over traditional volumes for both diagnostic and treatment prediction tasks, with interpretable weight maps providing insights into the local features driving model performance.

BackgroundMagnetoencephalography (MEG) is a unique technique in human neuroimaging combining high temporal resolution (millisecond or faster) with moderate spatial resolution (several millimeter). While many software packages for MEG data analysis exist, there is no pipeline developed for the specific purpose of enabling the automated analysis of very large, multi-site datasets. ResultsThe ENIGMA consortium was developed to enable large scale collaborations in the fields of neuroimaging and genetics. To facilitate ENIGMA MEG working group data analysis, we developed the ENIGMA MEG pipeline. The first ENIGMA MEG working group project involves spectral analysis of resting state MEG data, thus our current pipeline is designed to carry out that task. The goals of the ENIGMA MEG pipeline include ease of use, automated processing wherever possible, detailed logging and quality assurance (QA) features, the use of the brain imaging data structure (BIDS) format, anonymized output, and consistent processing across vendors. The pipeline is built using the MNE-Python framework and incorporates a re-trained version of the MEGnet deep neural network algorithm for automated artifact detection. QA tools are designed to enable high throughput evaluation of a large number of subject datasets. All software is open source and available on GitHub (https://github.com/nih-megcore/enigma_MEG). We used our pipeline to process data from three publicly available MEG cohorts, demonstrating its functionality and compatibility with large-scale processing. ConclusionsWhile the current ENIGMA pipeline is limited to resting state data and spectral analysis for the current working group project, the software is highly modularized, allowing straightforward extension to other analysis questions. Further development of the tool to enable connectivity and task-based MEG analysis are planned. The ENIGMA MEG pipeline represents an important first step to augment the existing arsenal of analysis tools, enabling multi-site, high throughput data analysis.

SignificanceAccurate and reproducible optode placement is crucial for obtaining high-quality fNIRS data in both individual and group-level neuroimaging studies. Conventional optode/probe montage design tools usually transform a probe layout defined in 2D Cartesian space onto a 3D head surface using a mass-spring model. Such mechanical transformation, combined with the indirect mapping between the 2D probe definition and the 3D target space, can introduce placement variations across different head surfaces and subjects. AimWe introduce NeuroCaptain v2, an open-source Blender-based add-on designed to enable interactive, anatomically guided optode design, registration, and cortical sensitivity visualization for fNIRS head-cap and probe creation. ApproachNeuroCaptain v2 enables researchers to add, move, and edit fNIRS sources and detectors directly over a 3D head surface mesh, defining anchored optode positions, as well as setting the stiffness of the springs between adjacent optodes. It then utilizes Blenders built-in physical simulation engine to relax the initial probe layout to satisfy the mechanical constraints. With the built-in mesh-based Monte Carlo (MMC) and diffusion-solver Redbird, NeuroCaptain v2 computes and renders 3D sensitivity maps to guide iterative optode adjustment. The resulting 3D optode layout is stored in the form of barycentric coordinates defined in a 10-20 landmark mesh, enabling consistent probe transfer across different head models. ResultsWe demonstrate interactive 3D montage design, cross-head-atlas probe registration, and cortical sensitivity visualization across multiple head geometries. Registering a probe across seven neurodevelopmental head atlases, the proposed anatomical-coordinate approach yields a mean per-optode standard deviation of 2.29 mm, a roughly 74% reduction in cross-subject placement variability compared to 8.68 mm using a conventional 2D-to-3D registration. ConclusionsNeuroCaptain v2 provides a reproducible, fully open-source workflow for fNIRS probe montage design that facilitates anatomically guided probe development and cross-subject registration directly in a three-dimensional anatomical environment.

Opfically pumped magnetometer-based magnetoencephalography (OPM-MEG) has recently emerged as a powerful neuroimaging approach in cognifive neuroscience, extending beyond the limitafions of convenfional cryogenic systems with greater experimental flexibility and wearable recording. Despite these advantages, standardised data analysis frameworks specifically tailored to OPM technology are sfill lacking, leading to variability in processing choices and reduced reproducibility across laboratories and hardware plafforms. We introduce OPM-FLUX, a comprehensive and fully documented end-to-end analysis pipeline developed for OPM-MEG data. The pipeline defines a clear sequence of preprocessing, noise suppression, arfifact handling, spectral analysis, evoked response analysis along with recommended parameter seftings. It also includes source reconstrucfion to idenfify where in the brain the signals originate. In addifion, OPM-FLUX supports mulfivariate paftern analysis (MVPA), enabling fime-resolved decoding of cognifive processes from sensor level data. OPM-FLUX is implemented in MNE-Python and distributed as interacfive Jupyter Notebooks that combine executable code with detailed methodological explanafions and graphical outputs. The pipeline further provides standardized reporfing templates and a data acquisifion Standard Operafing Procedure to facilitate preregistrafion, consistent documentafion, and standard pracfices across research sites. The workflow is demonstrated using openly available datasets acquired from both Cerca/QuSpin and FieldLine OPM systems during a visuospafial aftenfion paradigm that modulates alpha, beta, and gamma oscillafions and elicits event-related responses. By supporfing mulfiple OPM plafforms and promofing consistent methodological choices, OPM-FLUX enhances transparency, comparability, and replicafion in OPM-MEG research. The pipeline also serves as an educafional resource for students and researchers entering the field and is designed to evolve alongside ongoing technological and methodological advances in OPM-based brain imaging.

As neuroimaging analysis shifts toward large-scale, multi-site studies, managing the unwanted variability introduced by combining heterogeneous datasets has become a critical challenge. Although tools such as ComBat and its neuroimaging extensions are widely used to address this variability, they only permit the modeling of categorical site effects and cannot account for continuous sources of confounding, such as image quality, head motion, and acquisition parameters. We introduce ComCat, an extension of the ComBat framework that preserves biologically relevant covariates while removing the effects of categorical site indicators and continuous nuisance variables. The latter are modeled as smooth nonlinear functions via B-spline basis expansion. ComCat is applicable to a broad range of brain analysis tasks, including voxel- and surface-based morphometry, normative modeling, and machine learning-based prediction. To demonstrate its capabilities, we evaluated ComCat on brain age prediction across five datasets covering complementary multi-site harmonization scenarios: ON-Harmony (10 subjects x 6 scanners; n = 80); the Buchert traveling-phantom dataset (1 subject x 116 scanners; n = 531); the Tohoku single-scanner, varying-acquisition dataset (n = 121); MR-ART (148 subjects with varying motion levels); and an ABIDE subset comprising 229 control subjects and 208 individuals with autism spectrum disorder across 14 scanners. Using image quality measures derived from CAT12 as continuous nuisance variables, ComCat reduced the mean absolute error (MAE) in brain age prediction relative to ComBat-GAM in all five datasets, including the two scenarios where site information was unavailable or uninformative. In the ABIDE dataset, ComCat improved harmonization while preserving the difference between the control and ASD groups, demonstrating that scanner-related variance can be removed without affecting biologically meaningful signals. ComCat can operate with or without site labels and is agnostic to the source of image quality metrics.

Portable low-field MRI systems are a promising complement to conventional high-field systems, enabling broader access to MRI. However, correspondence in cortical thickness estimates between low- and high-field MRI in young people remains limited despite its importance for neurodevelopment and psychopathology. To evaluate how multiple low-field image processing approaches improve cortical thickness correspondence with high-field MRI in a large sample of young individuals, we collected ultra-low-field (64mT) and high-field (3T) MRI data from a community sample of young people. We applied deep learning-based image processing approaches (SynthSR v1.0, SynthSR v2.0, recon-all-clinical, and recon-any) to low-field data acquired across multiple sequences (T1- and T2-weighted) and orientations (axial, coronal, sagittal, and multi-orientation), with and without resampling and/or co-registration. We assessed global, lobar, and regional cortical thickness correspondence with 3T MRI measures using Pearson and intraclass correlations. We compared pipelines using Steigers Z-tests and Fishers Z-tests. A total of 150 individuals (mean age, 18.63{+/-}5.07; 80 female) were included. We observed the highest global correspondence with recon-all-clinical applied to coronal T1-weighted images (r=0.40, pFDR=2.6e-05). At the lobar and regional levels, multi-orientation T2-weighted images processed with recon-all-clinical showed the highest correspondence across the greatest number of regions (4/12 lobes; 13/68 regions). The highest correspondence and largest improvements were in frontal, cingulate, and temporal regions, including the right pars triangularis (r=0.52, pFDR=4.78e-11; Z=4.78, pFDR=4.25e-06), right caudal anterior cingulate (r=0.47, pFDR=3.83e-09; Z=5.46, pFDR=1.32e-07), and left parahippocampal (r=0.58, pFDR=2.98e-14; Z=5.17, pFDR=6.01e-07). We observed significantly improved cortical thickness correspondence in low-field MRI in young people. The recon-all-clinical pipeline yielded moderate correspondence, particularly in frontal, cingulate, and temporal regions. Our results highlight the potential of low-field MRI as an affordable and scalable approach for assessing cortical thickness in young people.

Macaque brain MRI is central to translational and comparative neuroscience, yet multi-site, longitudinal, and cross-species analyses are hindered by a lack of unified, automated structural processing tools. Existing pipelines, mostly adapted from human neuroimaging or restricted to fragmented steps, fail to provide robust surface-volume representations across heterogeneous acquisitions and developmental stages. Here we introduce MacaSurfer, a fully automated, containerized framework for unified surface-volume mapping of the macaque brain across the lifespan. MacaSurfer features components tailored for macaque anatomy: a tissue segmentation model, a tissue-guided bias-field correction method optimizing structural mapping from T1-weighted images alone, topology-aware surface reconstruction, and surface-aware volumetric registration. Validated on 1,346 imaging sessions from 965 macaques across 39 international sites (spanning 2 weeks to 23 years of age), MacaSurfer demonstrated exceptional anatomical consistency, test-retest precision, and robustness against image degradation. Leveraging MacaSurfer-derived morphometry, we established normative trajectories from 835 macaques, providing a standardized reference for downstream individualized deviation analysis. MacaSurfer is openly available with source code, containers, and pretrained models, offering a reproducible ecosystem to accelerate developmental, translational, and comparative neuroimaging.

Macaque MRI bridges non-invasive systems neuroscience with cellular and circuit-level mechanisms, but preprocessing remains fragmented across tools that are difficult to integrate, adapt to non-human primate acquisitions, and deploy reproducibly. We present Brainana, an automated, BIDS-compatible preprocessing framework for macaque neuroimaging. Brainana integrates structural and functional preprocessing, cortical surface reconstruction, quality control, transform tracking, and atlas projection within a containerized package, with cloud access for users without local compute. It incorporates macaque-trained deep learning models for brain extraction and tissue segmentation, conformation to standardize variable acquisitions, and surface reconstruction optimizations for macaque neuroanatomy. Across 23 imaging sites, Brainana processed data spanning heterogeneous scanners, protocols, species, and resolutions, yielding accurate anatomical correspondence across 130 monkeys, reliable native-space cortical surfaces, localized task-evoked activations, and reproducible brain-wide resting-state correlation structure. Brainana enables reproducible, scalable, and accessible macaque MRI preprocessing that supports cross-study comparison and multimodal integration across spatial scales, from neurons to networks.

Zero-shot learning from functional magnetic resonance imaging (fMRI) data offers a principled approach to decoding conceptual knowledge without requiring training examples for every target concept. The Semantic Output Code (SOC) framework, introduced by Palatucci et al. [2009], operationalises this idea through a two-stage pipeline: a regression-based mapping from voxel activations to a semantic feature space (the S map), followed by nearest-neighbour retrieval over a semantic knowledge base (the L map). Despite its foundational role in the field, no fully documented, open-source replication of this framework has been published on the original Mitchell et al. [2008] fMRI dataset. We present such a replication and extend it through a systematic evaluation of every major design choice in the pipeline. Using the official 25-verb co-occurrence feature space from Mitchell et al. [2008] and the correlation-stability voxel selection criterion, our pipeline achieves a mean pairwise 2-way forced-choice accuracy of 76.5% (SD = 4.9%, range: 70.0%-84.1%) across all nine subjects of the Mitchell dataset, within 0.5 percentage points of the published benchmark of 77%. We document and resolve a previously unreported evaluation artefact caused by a degenerate zero-vector knowledge base entry for one stimulus word (skyscraper), which suppressed accuracy by approximately 8 percentage points under the broken configuration. Sensitivity analyses across regularisation strength, voxel count, and knowledge base normalisation demonstrate that the pipeline is robust to hyperparameter choice within a broad operating range, with voxel count being the single most impactful factor. Substantial inter-subject variability is documented, with pairwise accuracy ranging from 70.0% (P9) to 84.1% (P1), a spread of 14.1 percentage points that exceeds the difference between our mean and the Mitchell benchmark. All code, the expanded 60-word knowledge base, and the complete evaluation pipeline are released as open-source software at https://github.com/Rashed525/fmri-zsl-pipeline.

High-resolution postmortem (ex vivo) magnetic resonance imaging enables detailed examination of brain anatomy at spatial scales not achievable in vivo and provides a unique opportunity to link morphometric measurements with the underlying pathology. Despite these advantages, robust computational tools for automated anatomical segmentation and cortical surface reconstruction remain limited, particularly in postmortem infant brains. Incomplete myelination, thinner cortical ribbons, small-scale neuroanatomy, as well as an evolving tissue contrast combined with fixation-induced signal alterations and variability in postmortem preparation make standard neuroimaging pipelines unusable for postmortem infant MRI. In this work, we introduce a one-of-its-kind multi-modal high-resolution postmortem infant MRI dataset and a unified computational framework that combines deep learning-based volumetric segmentation with surface-based cortical reconstruction and anatomical parcellation in native subject space resolution. To address the pronounced domain shift inherent to postmortem MRI, we develop a postmortem-specific synthetic data generation engine (PostSynth) that explicitly models fixation-driven postmortem imaging characteristics. In particular, we incorporate postmortem-specific altered gray-white matter contrast, laminar cortical intensity heterogeneity, specimen-specific bias fields, and background signal characteristics associated with immersion media: phenomena not typically observed in in vivo data or captured by generic contrast-agnostic synthesis methods. We benchmark our framework against a set of widely used contrast-agnostic and foundational brain segmentation models, demonstrating improved anatomical consistency and segmentation performance in high-resolution postmortem infant data. The code is publicly available as part of the purple-mri package.

Clinical Dementia Rating (CDR) scores are used to classify the cognitive state of patients and are provided within neuroimaging datasets. This is achieved through a standardized clinical assessment that evaluates participants cognitive and functional abilities in everyday life, after which they are given a score ranging from 0 to 3. Where 0 represents no signs of dementia and three represents severe dementia1. These scores are then used to track the progression of dementia over time2. This study explored if these CDR labels within the OASIS-1 dataset produced consistent volumetric separation across the hippocampus, amygdala, and cortex.

Real-time functional magnetic resonance imaging (fMRI) offers a powerful means of studying brain function adaptively, enabling experimental parameters to be updated dynamically in response to ongoing neural activity. However, current approaches remain limited by complex infrastructure requirements, bespoke implementations, and a lack of flexible frameworks for closed-loop neuroimaging, with many primarily focusing on neurofeedback experimental designs. Here we present AutoNeuro, an open-source framework for real-time fMRI acquisition, preprocessing, feature extraction, and adaptive experimental control. AutoNeuro connects directly to the MRI scanner, receiving reconstructed slices as soon as they become available, and streams them into a modular analysis pipeline designed for low-latency processing. Neural features are estimated at the temporal resolution of acquisition and are passed to a Bayesian optimisation agent that selects task conditions to maximise a user-defined objective function. Experimental conditions are represented within a bounded "experiment space", allowing heterogeneous conditions to be explored within a common coordinate system. We demonstrate AutoNeuro in a real-time fMRI experiment in which the system adaptively sampled task conditions to obtain a continuous map of brain response to the range of conditions contained within the experimental space. The system operated within the temporal constraints of real-time preprocessing and analysis, maintaining stable model estimates across iterations, converging on experimental conditions most relevant to the measured brain metric. These results establish AutoNeuro as a flexible platform for closed-loop neuroimaging, supporting hypothesis-driven optimisation as well as exploratory mapping of brain metrics across large experimental spaces.

Brain magnetic resonance imaging (MRI) is essential for diagnosis and neurodevelopmental research, but the high cost and infrastructure demands of high-field MRI limit its use to high-income settings. Ultra-low-field MRI scanners offer a more affordable and energy-efficient alternative, but their reduced resolution and signal-to-noise ratio restrict research and clinical utility, prompting the need for super-resolution techniques. Current super-resolution methods rely on either three anisotropic ultra-low-field scans acquired at different orientations (axial, coronal, sagittal) to reconstruct a higher-resolution image using multi-resolution registration (MRR) or the training of deep learning models using paired ultra-low- and high-field scans. Since acquiring three high-quality ultra-low-field scans is not always feasible, and paired high-field data may not be available, this study explores the efficacy of using a deep learning model to generate scans of MRR quality from a single ultra-low-field input scan. Results demonstrated significant enhancement in the quality of output scans, including improved image quality metrics, stronger tissue volume correlations, and greater Dice overlap of tissue segmentations. Generating higher-resolution brain scans from single ultra-low-field scans, without paired high-field data, reduces scanning time and further widens MRI accessibility in low- and middle-income countries. This approach also facilitates site-specific model training, which an exploratory external validation suggests may be necessary to address potential domain shifts across scanning sites.

We demonstrate that features extracted from structural MRI using un-CNN, an untrained convolutional neural network, achieve predictive performance comparable to or exceeding that of state-of-the-art pretrained foundation models across three structural MRI datasets and three downstream tasks. Un-CNN extends a classical 3D CNN architecture with multi-channel inputs, a hierarchical encoder with multi-scale feature aggregation, and covariance pooling. Untrained CNNs circumvent several key limitations of trained models, including high computational cost and memory requirements, the need to distribute large model weights, risks of data leakage, and challenges in reproducibility.