Thomas J. Madison

Heritability of regional subcortical brain volumes (rSBVs) quantifies the contribution of genetic variation to individual differences in subcortical brain morphology during neurodevelopment. rSBVs are highly heritable in adults but are not well characterized in adolescents. The Adolescent Brain Cognitive Development study (ABCD), collected across 21 US sites, provides data to characterize the heritability of subcortical structures in adolescence. However, in a multi-site multi-ancestry study, such as ABCD, site-specific effects may co-occur with genetic effects, which can bias heritability estimates. Existing methods that adjust for site effects require additional steps and can lead to inconsistent estimates. We propose a random-effect model-based MOM approach to estimate heritability that is a single-step asymptotically consistent estimator even when ancestries are differentially distributed across sites. Our approach performed well under a wide range of simulations. The proposed approach yielded heritability estimates similar to previous results derived from single-site studies and studies on different age groups, with the cerebellar cortex and hippocampus showing the highest estimated heritability ( > 50%).

Individualized resting-state functional magnetic resonance imaging (rs-fMRI) is increasingly used to guide neuromodulation target selection. However, clinical scans are often short and noisy, and standard pipelines for functional network identification do not provide information about confidence of network assignment. With limited data, unstable network assignments can misdirect stimulation toward off-target regions, making it critical to know which assignments can be trusted. We developed Precision Confidence Mapping (PCM), a bootstrap-based framework that makes this uncertainty explicit and actionable. PCM repeatedly resamples the time series and reruns network detection to estimate how consistently each vertex is assigned to a given network. The resulting confidence maps can be thresholded to exclude less stable regions. We evaluated PCM across scan durations from 5 to 70 minutes using positive predictive value (PPV) as the primary measure of network-assignment precision. PPV quantified the proportion of vertices assigned to a network that received the same label in an independent within-subject 70 minute reference map. Confidence thresholding markedly improved PPV across functional networks, with the largest gains for short scan durations. Compared with standard network assignment, PCM significantly increased agreement with this independent reference. Within-subject agreement remained greater than between-subject agreement, indicating that thresholding preserved individual-specific network topography. These precision gains came with modest reductions in reference-network coverage, particularly at shorter scan durations. This tradeoff may be acceptable for neuromodulation applications that prioritize minimizing off-network assignments. By adding a reliability layer to individualized mapping, PCM supports more cautious and precise neuromodulation targeting under real-world clinical scan constraints.

Functional MRI (fMRI) data are severely distorted by magnetic field (B0) inhomogeneities, which currently must be corrected using separately acquired field map data. However, changes in the head position of a participant across fMRI frames cause changes in the B0 field, preventing accurate correction of geometric distortions. Movement during field map acquisitions corrupts field maps, preventing distortion correction altogether. In this study, we use multi-echo (ME) fMRI data to dynamically sample and correct for magnetic field image distortions caused by head motion. Our distortion correction pipeline, MEDIC (Multi-Echo DIstortion Correction), leverages magnetic field inhomogeneity information found in the difference between echoes and uses it to correct for distortion on a frame-by-frame basis. Here, we demonstrate that MEDIC’s frame-wise distortion correction decreases the impact of head motion on resting-state functional connectivity (RSFC) maps and improves alignment to anatomy when compared with the prior gold standard approach (i.e., FSL TOPUP). Enhanced frame-wise distortion correction with MEDIC, without the requirement for field map collection, furthers the benefit of cutting-edge multi-echo fMRI imaging over single-echo fMRI.

Background: Existing evidence suggests cortical morphometric alterations occur in people with autism and ADHD. However, these findings remain tentative due to small sample sizes, heterogeneous imaging pipelines, varied statistical approaches, and limited harmonization across acquisition sites. Few studies have applied standardized processing to large, clinically enriched datasets or addressed site-related batch effects. Methods: We leveraged six large-scale brain imaging datasets (n = 9,647; male=5,835; female=3,812; ages 5–64 years), including 1,533 individuals with ADHD, 1,080 with autism spectrum disorder, and 7,034 matched controls. All imaging data were processed using the validated ABCD-HCP pipeline, with cortical parcellation into 360 regions based on the Human Connectome Project (HCP) atlas, and ComBat harmonization was applied to account for variability across 67 acquisition sites. Group-level differences in cortical thickness and sulcal curvature were examined with ANCOVAs, controlling for covariates and using Bonferroni correction for multiple comparisons. Results: Our analyses revealed distinct neuroanatomical signatures for both autism and ADHD. Individuals with autism exhibited regionally thinner cortex and curvature alterations particularly in the Cingulo-Opercular network. In contrast, individuals with ADHD displayed regionally thicker cortex, particularly in the default mode and somatomotor networks, alongside curvature differences. Control participants showed intermediate patterns, suggesting that autism and ADHD may represent diverging extremes of cortical maturation. Conclusions: Cortical thickness and curvature emerge as potential biomarkers that can advance understanding of neurodevelopmental conditions and disentangle heterogeneity across diagnostic groups. These findings highlight the value of harmonized, large-scale, standardized analyses for resolving inconsistencies in the literature.

Personalized functional brain developmental trajectories can be studied with Precision Functional Mapping (PFM). Our previous work has demonstrated that PFM can be achieved in infants despite rapid brain growth. However, even with extensive data collection (up to 1 hour of fMRI), the reliability and precision of these maps remain lower than those observed in youth and adults - particularly within subcortical structures. In this work we demonstrate the utility of high-field 7T MRI compared to 3T MRI for facilitating PFM in infants. We showcase data from multi-echo fMRI acquisitions in the same infants at both 7T and 3T and demonstrate that 7T imaging in infants is safe and feasible with our subject-specific safety workflow. Moreover, we demonstrate that the use of a higher magnetic field strength affords a spatial resolution more appropriately matched to infants’ smaller head and brain sizes, yielding notable improvements in data quality, especially for PFM. The increase in both spatial precision and reliability also suggests that 7T MRI can reduce the amount of data required for PFM. Last, we show how ultra-high field imaging can help us study the development of subcortical-to-cortical connectivity patterns, crucial for understanding brain development during this developmental window. 7T MRI is a promising new avenue for developmental cognitive neuroscience.

Understanding the mechanisms underlying mental illness requires disentangling the roles of genetic and environmental factors. The brain has been proposed as a key intermediate biomarker linking these factors, yet the extent to which genetics shapes adolescent brain function remains elusive. Using data from the ABCD study (5,247 unrelated, 330 dizygotic, 248 monozygotic twin participants), we estimated heritability of functional connectivity and topography using both SNP and twin data. SNP heritability, calculated using genetic correlation and the AdjHE-RE estimator, was low for connectivity (median = 2e-10% Gordon, 5.8e-7% probabilistic) and topography (max = 2%). Twin-based ACE models replicated prior literature (median = 8.9e-6% Gordon, 6% probabilistic connectivity, 27% topography). These results indicate that additive genetic effects have minimal influence on adolescent functional brain features, emphasizing the need to integrate environmental factors when studying the development of functional brain networks and their relevance to mental health.

Treatment-resistant depression (TRD) is a leading cause of premature death. For decades investigators have assessed the clinical efficacy of direct brain stimulation for TRD. Outcomes have been inconsistent due to imprecise brain targeting. Minimally invasive Personalized Adaptive Cortical Electro-stimulation (PACE) uses fMRI-based precision functional mapping (PFM) to target patient-specific network anomalies with brain surface electrodes. We utilized this novel technology in an n-of-1 study to treat a 44-year-old man (TRD-1) with over 30 years of severe unipolar depression. PFM revealed a 400% expansion in cortical area of the Salience Network (SN) and a 25% reduction in default mode (DMN) and frontoparietal (FP) networks compared to a group average controls. Stimulation paddles were implanted targeting patient-specific SN, DMN, and FP networks. Postoperative stimulation testing (within 24h) revealed immediate electrode-specific mood and cognition responses that matched the underlying functional networks. Stimulation parameters were iteratively optimized using ‘Bayes Tuning’ based on patient feedback. TRD-1’s suicidal ideations ceased within 7 weeks. Full remission of symptoms was achieved within 9 months and maintained at 30 months. PACE therapy appears to be a scalable, safe, and cost-effective approach to treat TRD.

Adaptively responding to salient stimuli in the environment is a fundamental feature of cognitive development in early life, which is enabled by the developing brain. Understanding individual variability in how the brain supports this fundamental process is essential for uncovering neurodevelopmental trajectories and potential neurodevelopmental risks. In the present study, we used a precision functional imaging approach to probe activation in response to salient auditory stimuli and its relation to brain functional networks in individual infants. A minimum of 60 minutes of fMRI BOLD data with an auditory oddball paradigm were collected in ten infants with a mean postmenstrual age of 48 weeks. Results demonstrate the feasibility of performing a precision functional imaging study to investigate individual specific responses to salient stimuli in infants. While responses to the auditory oddball were consistent between individuals in auditory processing areas, responses across the rest of the brain differed across individuals in their magnitude and shape. Individual specific response patterns appeared to be relatively stable and differed from other participant’s response patterns, despite fluctuations across runs. Commonalities and differences between individuals demonstrated in this sample contribute to our understanding of how the developing brain instantiates processing of salient stimuli. Our findings suggest that during early development, early unimodal processing is well conserved across individuals, however subsequent perceptual processing is still being personally defined. In this context, individual specific response patterns could be a promising target for biomarkers of normative brain and cognitive development.

The adoption of a standardized preprocessing workflow is vital for fostering community, sharing, and reproducibility. fMRIPrep has been a critical advancement towards this end, however, it is limited in its capacity to be applied to data across the lifespan, starting from infancy. Here, we introduce fMRIPrep Lifespan, an extension of fMRIPrep that extends the standardized processing from childhood to senescence to include neonatal, infant, and toddler structural and functional MRI data preprocessing. This effort involves a NiPreps integration of 1) a workflow akin to fMRIPrep optimized for MRI data in the first years of life (previously NiBabies) and 2) upstream enhancements to the entire NiPreps suite, including multi-echo data processing, modularization of workflow components, and convergence of processing with other popular workflows (ABCD-BIDS, Human Connectome Project Pipelines). Using data from the Baby Connectome Project (participants 1-43 months of age), we demonstrate that fMRIPrep Lifespan produces high-quality outputs across a wide age range. Moving forward, the scalable, modular infrastructure of fMRIPrep Lifespan will ensure adaptability to data from birth to old age while maintaining robust and reproducible frameworks for functional MRI research across the lifespan.

Heritability of regional subcortical brain volumes (rSBVs) describes the role of genetics in middle and inner brain development. rSBVs are highly heritable in adults but are not characterized well in adolescents. The Adolescent Brain Cognitive Development study (ABCD), taken over 22 US sites, provides data to characterize the heritability of subcortical structures in adolescence. In ABCD, site-specific effects co-occur with genetic effects which can bias heritability estimates. Existing methods adjusting for site effects require additional steps to adjust for site effects and can lead to inconsistent estimation. We propose a random-effect model-based method of moments approach that is a single step estimator and is a theoretically consistent estimator even when sites are imbalanced and performs well under simulations. We compare methods on rSBVs from ABCD. The proposed approach yielded heritability estimates similar to previous results derived from single-site studies. The cerebellum cortex and hippocampus were the most heritable regions ( > 50% ).