Irritability is a prevalent and impairing feature associated with autism, yet remains poorly understood, particularly in adults. Drawing heavily on insights translated from pediatric and transdiagnostic literatures, we propose that irritability in autistic individuals often reflects a psychophysiological stress or threat response, rooted in a vulnerable neurobiology (e.g., sensory sensitivities, intolerance of uncertainty), but may also stem from intrinsic neurobiological dysregulation independent of environmental triggers. The current treatment paradigm for autistic adults, largely extrapolated from pediatric antipsychotic trials, leaves these adults critically underserved due to a lack of evidence-based treatments, clinical trials, or validated tools to measure their internal experience. This viewpoint deconstructs irritability, differentiating its affective nature from aggression, and highlights heterogeneity across the lifespan and support needs. We critique the limitations of current assessment methods and recommend a shift toward a multi-modal strategy integrating self-report (when feasible) with objective, physiologically-informed tools (e.g., wearable biosensors) and nuanced observer reports. We argue that the field is poised for neuroscience-informed treatment innovation-including novel pharmacological agents and adapted psychosocial interventions-but is hampered by a lack of rigorous clinical trials in adults. Finally, we call for mechanistically driven trials to address the large unmet burden of inadequately treated irritability, ultimately improving the quality of life for autistic adults and their families/caregivers. Irritability in autistic adults is a major crisis, but most treatments are decades old and were only tested on children. Drawing on evidence primarily from youth and related conditions, our review explains that irritability is often a physical and emotional stress response to things like sensory overload or trauma, not just a “behavior problem.” We call for urgent research into new, adult‐focused treatments and tools, like wearable sensors, to directly address the root causes of this distress.
This paper reports a genetic identification task using 3D convolutional neural network (3D-CNN) models applied to a proprietary 3D magnetic resonance imaging (MRI) dataset of patients with lissencephaly. Lissencephaly is a neuronal migration disorder caused by genetic mutations or deletions in which specific causative genes result in distinct morphological alterations in brain structure. The objective of this study was to identify causative genes through image classification by analysing three-dimensional structural features of brain MRI using deep learning. In our experiments, we extended representative CNN architectures to handle three-dimensional inputs and performed three-class classification targeting the primary causative genes, LIS1 and DCX, along with a category for other genetic variations. Our results demonstrated that the 3D-ResNet18 model achieved a mean classification accuracy of over 78%. Furthermore, to enhance the precision for primary genes, we introduced a decision-making process based on prediction probability thresholds. This approach yielded an average precision improvement of 4.67% for DCX and 4.84% for LIS1 across all the evaluated models.
To examine how three dimensions of the Multidimensional Attitude toward Ambiguity Scale (MAAS)-Discomfort with Ambiguity, Absolutism, and Need for Complexity-relate to resting-state functional connectivity, conceptually replicating and extending the work of Liu et al. (2023) in a Japanese sample. Liu et al. (2023) reported that higher ambiguity tolerance was associated with stronger connectivity in integration and control networks, whereas lower ambiguity tolerance was associated with stronger connectivity in threat- and error-monitoring circuits. Of the three MAAS dimensions, Need for Complexity was the one most closely aligned with their measure. Thirty-nine participants underwent resting-state MRI and completed the MAAS. Region-of-interest (ROI)-to-ROI analyses were used to test the associations between each MAAS dimension and its hypothesized connectivity pair, controlling for age, sex, and head motion. No MAAS dimension was significantly associated with its corresponding connectivity pair. Effect sizes were negligible, although the Need for Complexity showed a small zero-order correlation with inferior parietal lobule-middle cingulate cortex connectivity. These findings contrast with earlier reports using unidimensional measures, suggesting that previously observed neural correlates may not map directly onto specific MAAS dimensions. Larger, well-powered, cross-cultural studies are needed.
Autism spectrum disorder (ASD) and attention-deficit/hyperactivity disorder (ADHD) are common neurodevelopmental disorders (NDDs) in children and often co-occur (ASD + ADHD), complicating the diagnosis. The diagnostic process is lengthy and subjective, relying heavily on expert knowledge, which limits accessibility. Electroencephalography (EEG) offers potential as a biomarker but requires skilled technicians for measurements and can be stressful for children with NDDs. This study aimed to provide a more accessible diagnostic support tool. We used a portable EEG device with a low participant burden and a deep learning model to distinguish between the typical development group (TD) and the NDD group, comprising children with ASD, ADHD, and ASD + ADHD. Resting-state EEG data were recorded for 5 min using the portable HARU-2 device, which features three channels placed on the forehead of 163 participants (87 TD, 76 NDD). A deep learning model combining a one-dimensional convolutional neural network and a transformer encoder was developed to analyze the EEG data. In 5-fold cross-validation, the model achieved an area under the curve (AUC) of 0.713 and a balanced accuracy (bACC) of 67.5% for classifying NDD and TD. Exploratory evaluation of out-of-fold predictions stratified by clinical phenotype showed an AUC of 0.793 and bACC of 75.0% for ASD vs. TD, 0.817 and 79.5% for ADHD vs. TD, and 0.667 and 62.7% for ASD + ADHD vs. TD. These findings suggest that portable EEG devices combined with deep learning models may serve as accessible adjunctive tools for NDD screening in children.
A high proportion of individuals with major depressive disorder (MDD) develop recurrent episodes. A better understanding and the identification of markers of recurrence risk are needed to develop prophylactic treatments and to support treatment decisions. Electroencephalography (EEG) is a scalable approach to identifying potential markers. Alpha band (8-13Hz) power and frontal alpha asymmetry (FAA) have been implicated in the pathophysiology of depression, however prospective studies of recurrence risk are sparse. Here, we investigated frontal alpha power and FAA in remitted MDD patients, to assess their potential as predictors of future depressive episodes. In total, 84 participants (n = 53 medication-free remitted MDD, n = 31 control participants with no family history of MDD) completed resting EEG testing. Remitted MDD patients were followed up clinically to determine recurrence over 14 months. Resting frontal alpha power and FAA were compared between groups (controls, stable remission, subclinical symptoms, recurrent episode) and binary logistic regression models were used to predict recurrence risk in remitted MDD patients. Patients who developed a recurring episode showed reduced frontal alpha power at baseline compared to controls and patients who remained in stable remission. No group effect was found for FAA. When combining clinical predictors with frontal alpha power, prediction accuracy for recurring episodes was 88%. This study demonstrates that reduced frontal alpha power has the potential to be further developed as a marker for recurrence risk in MDD, and could contribute to recurrence risk prediction models. This calls for larger studies using cross-validated prediction modelling techniques.
Functional magnetic resonance imaging (fMRI) studies have demonstrated music-induced activation of the blood-oxygen-level-dependent (BOLD) signal across brain networks associated with auditory perception, motor control, and emotion. However, BOLD-fMRI reflects vascular responses that may not fully capture underlying neural activity. Here, we used simultaneous [18F]fluorodeoxyglucose (FDG) functional positron emission tomography (fPET) and fMRI to examine glucose metabolism closely linked to neural activity, alongside hemodynamic responses during pleasurable music listening. Thirty-five female participants listened to self-selected pleasurable music and control stimuli while undergoing 90-minute PET-MRI scans. In fPET, the music > control contrast revealed music-evoked increase in glucose consumption in auditory and motor cortices, as well as reward-related regions, including the nucleus accumbens (NAc), caudate, insula, and orbitofrontal cortex. The fPET and fMRI results showed substantial overlap though some discrepancies were also observed. Notably, the NAc exhibited increased glucose consumption in fPET but showed no activation in fMRI. Conversely, deactivation of the default mode network during music processing was only observed with fMRI. These results highlight the complementary nature of neurometabolic and neurovascular processes and offer novel insights into their dynamics during the processing of aesthetic rewards.
We recently suggested that when visual objects move towards each other, it may involve a trajectory prediction and sensory checking at the level of milliseconds. We examined this possibility by exploring the consequences of a sub-threshold manipulation of squares trajectories on a newly found illusion, i.e., the perception of a gap at the time of the collision. We first examined the consequences of the manipulation on the conscious illusion. In a more exploratory way, we studied the EEG correlates of the manipulation, as well as the impact of transcranial magnetic stimulation (TMS) on the cerebellum (right Crus I/II). The TMS was a typical intermittent theta-burst stimulation, but only one sequence of around 3 min, compared with a placebo stimulation. All participants participated to two TMS sessions, in the sham-verum or in the verum-sham order. The trajectory manipulation occurred within less than 60 ms before the collision and reliably decreased the illusion rate. The results suggest that the prediction is temporarily stopped after the trajectory change. The illusion was accompanied by late positive ERP signals in CP4. The amplitude of these signals was modulated by TMS verum on the cerebellum, but only in the sham-verum order. As a whole the results are consistent with automatic verifications of the trajectory regularity at the millisecond level.
Dynamic deuterium metabolic imaging (DMI) enables time-resolved mapping of cerebral glucose metabolism in vivo, yet its intrinsically low SNR often renders voxel-wise metabolite quantification unstable-particularly at early repetitions. Low-rank denoising is widely used in MR spectroscopic imaging (MRSI)/DMI to improve robustness, but very low-SNR regimes and dynamic studies remain challenging. Self-supervised learning is promising for DMI/MRSI denoising, but its performance depends strongly on the representation domain and the noise assumptions underlying the training objective Here, we study a pragmatic denoising pipeline for dynamic DMI/MRSI that combines a mild low-rank stabilization with self-supervised learning in the spectral-temporal (f×T) domain. Exploiting metabolite-specific spectral structure and redundancy across repeated measurements, our approach relies on two mild assumptions: (i) additive, approximately zero-mean noise and (ii) approximate noise independence across repeated acquisitions. These conditions are expected to be satisfied across essentially any reconstruction and preprocessing pipeline. Our results indicate that the f×T domain is effective both with spatially correlated and approximately uncorrelated noise. We evaluate the approach in simulations and in vivo dynamic DMI data from healthy volunteers (n=6) and a brain tumor patient. The proposed pipeline improves the robustness of time-resolved metabolite estimates and increases LCModel fit stability relative to a state-of-the-art low-rank baseline (tMPPCA), with the largest gains for weak metabolites and early low-SNR repetitions. Together, this enables more reliable dynamic metabolite mapping in low-SNR regimes.
We examined the potential of combining EEG signals from multiple individuals to identify critical events in a team task. In this study two subjects played a video game in which they had complementary roles, one player serving as a Bait to distract 5 enemy fortress and the other serving as a Shooter to destroy the fortress. Twenty-one pairs of subjects were analyzed. Critical events, destruction of the fortress and deaths of each player, evoked distinguishable P300-like responses from both players. Fortress kills could be best identified by combining the two EEG signals, while deaths could be best identified by focusing on the response of the player who died. Hidden semi-Markov models (HSMMs) achieved good identification of the events by combining information about the temporal distribution of these critical events with the conditional probability of the EEG activity. These findings indicate that we can track and improve by adaptively merging or selecting the signals from different team members.
Selective auditory attention enables listeners to focus on a target speaker while suppressing concurrent sounds. To assess the ecological validity of virtual reality for studying spatial attention, the present study investigated cognitive and neural mechanisms in both real and virtual laboratory environments. Participants completed a selective spatial attention task in both real and virtual environments under auditory-only and audio-visual conditions while EEG was recorded. Speech stimuli were presented from different locations in both azimuth and depth. In each trial, two simultaneous speech stimuli originated from different locations. Each stimulus included a target or distractor word (plus visual patterns in the audio-visual condition) followed by a numeral. Participants indicated whether the numeral at the target location was odd or even, ignoring the competing numeral. Performance was slower and less accurate in the virtual environment, with auditory-only stimulation, and in depth-only trials, suggesting slower accumulation of sensory evidence, as confirmed by drift-diffusion modeling. Early sensory ERPs (P1, N1) were largely stable across conditions, whereas P2 amplitudes were modulated by environment, stimulation, and spatial configuration. Spatial configuration also strongly influenced amplitudes of the contingent negative variation, which were largest with azimuth-only stimuli. Importantly, these effects were consistent across environments and largely independent of visual cues and significant cross-environment correlations were observed for all behavioral and neurophysiological measures. These findings suggest that, despite additional perceptual and cognitive demands indicated by reduced performance and slower evidence accumulation, virtual reality captures key mechanisms of selective attention.
Growing evidence implicates self-blame-related neural networks in the pathophysiology of major depressive disorder (MDD). fMRI neurofeedback is an emerging technology with the potential to deliver interventions targeting the neural substrates of self-blame, but its feasibility and potential mechanism in current MDD remain unresolved. The current pilot study employed a single-session, exploratory neurofeedback trial harnessing the subgenual cingulate cortex (SCC) BOLD activity as the sole training target. Two active interventions were compared. In Intervention A (n = 10), participants were encouraged during neurofeedback to upregulate their SCC activity during a 'guilt' task and downregulate it during an 'indignation' task. In Intervention B (n = 10), participants were encouraged to do the opposite. Clinical scores improved significantly across interventions, although no significant intervention differences were detected. Neurofeedback performance was variable across participants and conditions, with significant group-level regulation in the intended direction observed only for the indignation condition in Intervention B. A whole-brain analysis using a uniform preprocessing pipeline revealed no clusters surviving correction for multiple comparisons. The results support the feasibility of the protocol and suggest that engagement with negative autobiographical memories during SCC-oriented neurofeedback can be delivered safely in this small sample. Group-level target engagement was inconsistent, but the observation that SCC regulation was most achievable during indignation tentatively suggests this region may be more functionally heterogeneous for causal agency representations than commonly assumed. The absence of significant whole-brain effects under a uniform pipeline indicates that any neural changes were not robust at the group level in this small sample. In summary, the study provides preliminary feasibility and safety data, alongside estimates of variance, to inform more adequately powered investigations into SCC-oriented neurofeedback for depression.
Pituitary adenomas represent one of the most common intracranial tumors, and cavernous sinus invasion (CSI remains a major challenge for surgical management. Although the Knosp grading system provides a widely used radiological framework, its subjective nature and inter-observer variability limit diagnostic reliability. In recent years, advanced computational methods have been investigated to improve the preoperative prediction of invasion. This review synthesizes current evidence on the use of radiomics, machine learning (ML), and deep learning (DL) approaches in the detection and assessment of CSI in pituitary adenomas, with particular emphasis on their comparative performance against traditional imaging methods. Studies employing MRI-based radiomic feature extraction, ML classifiers, and convolutional neural networks were analyzed. Reported models commonly incorporated intensity, texture, and shape descriptors, or applied end-to-end DL architectures for automated prediction. Performance metrics such as accuracy, sensitivity, specificity, AUC, and Dice similarity coefficients were compared across studies, with Knosp grade serving as a frequent benchmark. Evidence suggests that ML and DL models consistently outperform conventional MRI interpretation in predicting CSI. Radiomics pipelines integrating quantitative imaging features with clinical variables achieved high diagnostic accuracy, while CNN-based models trained on contrast-enhanced MRI often exceeded AUC values of 0.85. Furthermore, automated segmentation frameworks demonstrated reliable delineation of tumor boundaries, facilitating improved assessment of invasive behavior. Despite promising outcomes, limitations such as small sample sizes, single-center designs, and lack of external validation restrict broad clinical adoption. Radiomics and AI-driven approaches show substantial potential for enhancing preoperative evaluation of pituitary adenomas with CSI. Standardized imaging protocols, multicenter collaborations, and transparent model validation are essential for future integration into neurosurgical decision-making.
Hypertrophic olivary degeneration (HOD) is a rare neurological condition due to hypertrophy of the inferior olivary nucleus (ION), usually due to the disruption of the Guillain-Mollaret triangle (GMT). Here, we studied a patient with primary progressive apraxia of speech (PPAOS), a neurodegenerative disease condition of impaired motor speech production, who developed HOD and Parkinsonian features later in her disease course. We examined the patient's disease course and evaluated various clinical, pathological, and neuroimaging variables. The patient developed motor speech problems at the onset and was diagnosed with PPAOS on her first visit. Over time, she developed features of an atypical Parkinsonian disorder, but she never developed palatal or dentatorubral tremors or ocular myoclonus. MRI scans were performed yearly at each visit according to the research protocol. During her sixth visit, changes in the left ION were first observed as hyperintensity on the T2-weighted MRI consistent with HOD. Using advanced neuroimaging techniques, we identified decreased fractional anisotropy (FA) and increased mean diffusivity (MD) in white matter tracts to and from the ION, supporting the diagnosis of HOD. A pathological diagnosis of progressive supranuclear palsy was rendered at autopsy. The findings from this case study demonstrate that ION degeneration and HOD can be a late feature of PPAOS, even in the absence of associated clinical signs and symptoms.
The eye is often described as a window to the brain and may offer a non-invasive opportunity for investigating neurodegenerative changes. Although cerebral amyloid-β (Aβ) accumulation is a well-established hallmark of Alzheimer's disease (AD), the potential of ocular regions to capture aspects of this pathology in positron emission tomography (PET) remains underexplored. Our primary objective is to conduct a radiomics-SUVR analysis to investigate the relationship between ocular regions of interest (ROI) extracted from PET imaging and regional SUVR values obtained from six brain PET regions. Our secondary objective is to address domain shifts arising from heterogeneous data distributions across imaging sites during ocular segmentation. This study included 228 participants from Chonnam National University Hospital (CNUH) and an external validation cohort of 50 participants from ADNI-4, all with paired 18F-florbetaben PET and 3D T2-weighted MRI. For the segmentation task, we propose HCA-Net, a Hierarchical Cross-Attention Network that integrates co-registered PET and MRI using a multi-scale fusion strategy, allowing the model to better handle domain shifts in cross-site imaging conditions. For the radiomics-SUVR analysis, the segmented ocular ROI was classified into Aβ⁺ and Aβ⁻ groups using a CNN model. Radiomic features were then extracted and compared with corresponding regional brain SUVRs using statistical methods. While several features showed group differences at an uncorrected threshold, only one feature remained significant after correction for multiple comparisons highlighting the need for further validation. Correlation analyses identified modest associations between ocular radiomics and regional brain SUVRs, highlighting the influence of global amyloid burden on these relationships. Overall, our study presents a multimodal cross-site segmentation framework and provides preliminary evidence of neuro-ocular associations in AD through 18F-FBB PET imaging, highlighting the need for further validation.
Tumor segmentation involves the identification and definition of tumors in medical images, which is crucial for cancer diagnosis and treatment as it enables accurate measurement of the size and shape of the tumor. This can be done manually or automatically with computer algorithms. Automated segmentation reduces the time and effort of manual methods while enhancing accuracy and stability. This article uses BraTS2021 data for 3D brain tumor segmentation. Initially, an N4 BiasField Correction Filter preprocesses MRI data. An improved V-Net is used for segmenting enhanced tumors, whole tumors, and tumor cores. The V-Net processes image data in three dimensions, with enhancements in the decoder and encoder sections, improving overall performance. The encoder benefits from residual and dilated convolution layer, while the Attention Gate enhances the decoder. In the encoder, the dilated convolution layer enhances performance, while residual layers ensure that adding more layers doesn't harm the network. All four MRI types are input into the 3D improved V-Net, achieving Dice coefficients of 87%, 81.2%, and 74.43% for whole tumors, tumor cores, and enhanced tumors respectively. Results indicate this method effectively segments 3D brain MRI images.
In Parkinson's disease (PD), increased amplitude of high-frequency oscillations (HFO) has been confirmed to be coupled with the beta oscillations phase, resulting in increased phase-amplitude coupling (PAC) in the subthalamic nucleus (STN). This pathological coupling correlates with the severity of motor symptoms and is known to be modulated by therapeutic interventions. For example, dopaminergic medications and deep brain stimulation (DBS) were previously shown to reduce PAC, which changes in magnitude between states of rest and movement. However, PAC alterations during kinetic and static movements in the presence and absence of medications remain to be determined. Furthermore, there is little evidence on the relationship between PAC and clinical symptoms of PD. In this study, we investigated these two issues. We analyzed a publicly available dataset, which contained STN local field potential registrations (n = 20) and concurrent limb electromyography during resting, static, and kinetic movements with and without levodopa intake. We calculated the PAC within the STN during different conditions, and the PAC between the limb tremor and the ipsilateral LFP oscillations. Beta-HFO STN PAC was increased during the medication off period, especially during kinetic movement and was highly correlated with bradykinesia and negatively correlated with tremor. We observed no interaction between tremor frequency oscillation phase on electromyography and beta or HFO oscillation amplitude within the STN in either of the conditions. These results point out that beta-HFO PAC provides complementary information to beta power for electrophysiological localization during DBS implantation and during adaptive DBS.
Multiple Sclerosis (MS) disrupts white matter (WM) tract organization, affecting brain networks and neurological function. Current methods lack the capacity to directly quantify tract- and network specific myelin-sensitive measures with clinical outcomes. This cross-sectional study (122 MS patients, 72 females, median age 48.5 years, median EDSS 3.0) applied multi-compartment Myelin Streamline Decomposition (MySD), an advanced MRI technique for assessing myelin-sensitive measures of WM bundles and networks in the presence of focal lesions. Patients underwent 3T MRI with magnetization transfer (MT)-weighted and multi-shell diffusion imaging. These data enabled multi-compartment MySD reconstruction to generate myelin-sensitive tract measures for the corticospinal tract (CST) and cingulum bundle (CB) and MVF-derived myelin-sensitive network properties. Associations were tested between tract-specific measures and five network measures with disability (Expanded Disability Status Scale (EDSS)), information processing speed (Symbol Digit Modalities Test) and neuroaxonal damage (serum neurofilament light chain), adjusting for covariables. Primary FDR-corrected analyses were considered for the left/right CST with EDSS and left/right CB with SDMT. All remaining analyses were considered exploratory. While primary CST-EDSS associations did not reach significance, a subgroup analysis revealed an association between myelin-sensitive tract measure of the left CST with SDMT z-scores (β = 0.57, p = 0.001, R2 Adj = 0.21). Sensitivity analysis demonstrated correlations between myelin-sensitive tract measure of the left CST with mean strength (ρ = 0.38, p < 0.001) and efficiency (ρ = 0.33, p < 0.001). Lower myelin-sensitive tract measures of the CST related to poorer cognitive performance. Multi-compartment MySD enables exploration of myelin-sensitive proxies and clinical outcomes in MS.
[This corrects the article DOI: 10.1016/j.ynirp.2026.100336.].
In the European Union (EU), donanemab is indicated in adults with early symptomatic Alzheimer's disease who are apolipoprotein E ε4 non-carriers or heterozygotes. Among these, patients without superficial siderosis at baseline, uncontrolled hypertension, or anticoagulant use are eligible. To assess efficacy and safety of donanemab in the EU-eligible population. A post-hoc conservative hybrid imputation method was implemented for clinical efficacy analyses during the TRAILBLAZER-ALZ 2 placebo-controlled period. In the 78-week long-term extension (LTE) participants in the early-start (randomised to donanemab) and delayed-start (randomised to placebo with donanemab initiation during the LTE) groups were compared to a propensity-weighted external control. Participants were switched to placebo after meeting amyloid-based treatment course completion criteria. By 76 weeks, donanemab-treated participants in the EU-eligible population had a mean Clinical Dementia Rating Scale (CDR)-Sum of Boxes change from baseline difference from placebo of -0.7 points (95% confidence interval, -1.0, -0.4) and a 40.3% lower risk of disease progression to the next stage (per CDR-Global score). Treatment benefit increased over 154 weeks for non-carriers and heterozygotes, including those meeting treatment course completion criteria by 52 or 76 weeks. In the placebo-controlled period, 119 (19.5%) and 49 (8.0%) donanemab-treated eligible participants experienced amyloid-related imaging abnormalities-edema/effusion and infusion-related reactions, respectively. Safety findings were similar among donanemab-treated participants in the placebo-controlled period and LTE delayed-start group. Consistent with previous TRAILBLAZER-ALZ 2 and LTE findings, donanemab significantly slowed disease progression compared to controls with a manageable safety profile in non-carriers and heterozygotes.
Complex congenital heart disease (CHD) is associated with reduced brain volumes, but little is known about the brain developmental trajectory in CHD beyond childhood, which is a critical period for brain maturation. This study reports alterations in brain volumes from a large cross-sectional dataset of patients with CHD and controls, with an age range from childhood to young adulthood. Patients and controls underwent 3 T cerebral MRI and overlapping cognitive assessments. Images were processed using Freesurfer 5.3. The dataset comprised 311 participants, 128 CHD and 183 controls aged between 9 and 32 years (male: 52.1%). Associations between the total brain and grey matter, white matter, and cerebrospinal fluid (CSF) volumes and age, sex, group (CHD vs. controls) and maternal education were analysed using linear mixed models. Global and total grey/white matter volumes were smaller in patients with CHD compared to controls (p < 0.001), whereas CSF volumes did not differ significantly between groups (p = 0.23). No significant interaction between the group, sex and age was found. Larger global brain volume was associated with higher maternal education (p < 0.001) and higher IQ (p < 0.001). Consistent lower brain volume in CHD than in controls throughout childhood and young adulthood suggests that there is no convergence towards healthy peers in CHD brain volumes over time. Functional correlates of smaller brain volumes underscore the importance of longitudinal studies in better understanding the evolution and determinants of impaired brain development in CHD populations.