Simon B. Eickhoff

With the rapid advancement of artificial intelligence (AI) technologies, growing interest has emerged in its potential application in pediatric healthcare. Understanding the factors influencing AI acceptability among parents and clinical staff is an important prerequisite for facilitating its integration in clinical routine. To date, these factors have not been systematically assessed across different stakeholder groups. Here we investigated AI acceptability among parents (first cohort n = 198; second cohort: n = 79) and pediatric health care professionals (n = 33) across different disease, diagnosis or treatment scenarios. The effects of demographic variables and contextual predictors, such as data privacy, AI knowledge and perceived disease severity, on willingness to use AI were evaluated. More liberal data privacy was associated with reduced willingness to use AI ( p = < .001). A higher perceived disease severity was linked to higher willingness to use AI ( β = .10, p = .013) in the second parent group. When the recommendations of AI and clinicians conflicted, parents were more likely to choose AI over the judgment of clinicians in treatment compared to diagnosis scenarios. Healthcare professionals showed similar patterns but additionally weighted perceived disease severity when resolving disagreements. These findings highlight the importance of addressing stakeholder concerns regarding disease specific customization, accuracy, data privacy and accountability when introducing AI into pediatric healthcare settings.

Background: Hormonal fluctuations during the menstrual cycle (and their suppression) are accompanied by altered mood, cognition, and sleep, yet their relationship to subjective sleep quality and glymphatic function - a sleep- dependent brain waste-clearance pathway that involves perivascular spaces (PVS) - is unclear. Methods: We examined daily oestradiol, progesterone, luteinizing hormone (LH), follicle-stimulating hormone (FSH), subjective sleep quality, and MRI-derived PVS volume (PVSV) in a single healthy woman across two 30-day periods: during her physiological menstrual cycle and, one year later, under oral contraceptive pill (OCP) use. We employed linear regression models to examine whether hormones are associated with sleep quality and PVSV, and whether they moderate the sleep quality–PVSV association during the physiological menstrual cycle. Subsequently, we conducted permutation-based Spearman rank correlations to assess whether day-to-day changes in sleep quality were associated with corresponding changes in PVSV across the physiological menstrual cycle and OCP use. False discovery rate (FDR) correction was used to control for multiple comparisons. Findings: Higher LH levels were associated with poorer sleep quality during the physiological cycle (b=2·15, 95% CI 0·64 to 3·67, p _FDR =0·028), while FSH moderated the sleep quality–PVSV association (b=−92·98, 95% CI −160·15 to −25·81, p _FDR =0·034), suggesting that lower FSH was associated with lower PVSV, whereas this association reversed at elevated FSH levels characteristic of the periovulatory window. During the physiological menstrual cycle, day-to-day decreases in sleep quality were associated with synchronous increases in PVSV, driven by centrum semiovale PVSV (ρ=0·55, 95% CI 0·30 to 0·67, p _perm =0·0095). This association was absent during OCP use and differed between hormonal milieus (Δρ=0·72, p=0·040). Interpretation: Ovulation is characterized by poor sleep quality and reduced PVSV, potentially reflecting noradrenergic modulation. The association between sleep quality and PVSV day-to-day changes is shaped by the hormonal milieu, being absent under OCP. PVSV increase associated with poor sleep may reflect elevated sleep pressure promoting vasomotor-mediated cerebrospinal fluid inflow. Funding: None

Background: The brain age gap (BAG), the difference between neuroimaging-predicted and chronological age, captures inter-individual variation in brain aging. Although sensitive to Alzheimer’s disease (AD) pathology, its longitudinal patterns across the clinical AD continuum and prognostic relevance remain unclear. Methods: 577 participants from the DELCODE cohort (>2,100 MRI scans) were analysed: healthy controls individuals (HC, N=202), and patients with subjective cognitive decline (SCD, N=248), mild cognitive impairment (N=93), and AD dementia (N=34). All underwent structural MRI, amyloid (Aβ _42/40 ) and phosphorylated tau181 assessment, and lifestyle-related dementia risk profiling (LIBRA). BAG was derived using brainageR. Associations with baseline cognition, cognitive decline, and clinical progression (up to eight years) were examined using mixed-effects and Cox models. Mediation analyses tested whether BAG accounted for LIBRA-cognition associations. Biomarker-related and clinical findings were replicated in ADNI (N=461). Findings: BAG showed excellent short-term reliability, increased stepwise across the clinical spectrum and was elevated in amyloid-positive SCD, but not in asymptomatic amyloid-positive HC. Longitudinal BAG increases were strongest in amyloid- and tau-positive participants (Aβ+T+). Higher BAG was associated with poorer baseline cognition and predicted cognitive decline, with strongest effects in Aβ+T+. All main findings replicated in ADNI. BAG was associated with LIBRA only in biomarker-negative participants and partly mediated associations with cognitive outcomes in DELCODE. Interpretation: BAG is a reliable non-invasive marker of structural brain health sensitive to AD pathology and to modifiable AD risk. Detectable divergence prior to objective cognitive impairment supports its relevance for early risk stratification and prevention-oriented research. Funding: Helmholtz AI Cooperation Unit (ZT-I-PF-5-163).

BACKGROUND Smartwatches are increasingly used for decentralized data collection in clinical research, but the everyday-life settings that make these data attractive also introduce variability. Before a smartwatch-derived feature can support clinical monitoring, its reproducibility must be established: features with low test-retest reliability weaken associations with clinical outcomes and potentially generate non-actionable signals. Reliability is expected to vary by feature type, aggregation window, and data availability, but has not been systematically screened in a clinical cohort. OBJECTIVE This study aimed to evaluate the test-retest reliability of candidate smartwatch-derived features for longitudinal monitoring in adults with advanced cancer and in healthy controls, and to determine how reliability depends on the temporal aggregation window. METHODS In a prospective single-centre observational cohort study, we analysed 8 weeks of Garmin Vivosmart 5 sensor data from 60 adults with advanced cancer, and 20 healthy controls. Out of 80 participants, 77 contributed usable smartwatch data. We examined 35 daily features across seven domains: heart rate variability, heart rate, respiration, oxygen saturation, sleep, activity, and smartwatch-derived stress. Test-retest reliability was quantified as the intraclass correlation coefficient (ICC(2,1)) across adjacent non-overlapping 1-, 3-, and 7-day windows, with 95% CIs from subject-level bootstrap resampling (10,000 resamples). Between-group and therapy-centred contrasts used permutation testing with Benjamini-Hochberg false discovery rate correction. RESULTS Reliability improved with longer aggregation windows in both cohorts. Between 1-day and 7-day windows, median ICC(2,1) increased from 0.53 to 0.73 in controls and from 0.66 to 0.79 in patients. The number of features reaching good-to-excellent reliability, defined as ICC(2,1)≥0.75, increased from 5 of 35 to 15 of 34 in controls and from 12 of 35 to 26 of 35 in patients. Heart rate and heart rate variability features were the most reliable, with 4 of 11 reaching weekly ICC(2,1)≥0.90 in both cohorts. Activity, respiration, sleep and oxygen-saturation features were more sensitive to aggregation window and data availability, showing larger gains from daily to weekly aggregation (e.g., step count ICC increased from 0.30 to 0.73 in controls). Weekly reliability did not differ significantly between patients and controls (median ΔICC=0.036, no feature survived FDR correction, all q>0.05). No feature showed a significant change in reliability around therapy (median ΔICC=0.016, all q>0.05). CONCLUSIONS Weekly aggregation improved the reliability of many smartwatch-derived features, but reliability remained feature specific. Heart rate and heart rate-variability features were consistently reliable, whereas selected sleep and oxygen-saturation features displayed only moderate reliability across all aggregation windows. Reliability was comparable across cohorts and stable around therapy, indicating that feature-wise estimates are transferable across these clinical contexts. Feature-level reliability screening is a prerequisite before smartwatch-derived measures are used in clinical monitoring.

Handgrip strength (HGS) is a significant biomarker for overall health, offering a simple, cost-effective method for assessing muscle function. Lower HGS is linked to higher mortality, functional decline, cognitive impairments, and chronic diseases. Considering the influence of anthropometrics and demographics on HGS, this study aims to develop a corrected HGS score using machine learning (ML) models to enhance its utility in understanding brain health and disease. Using UK Biobank data, sex-specific ML models were developed to predict HGS based on three anthropometric variables and age. A novel biomarker, ∆HGS, was introduced as the difference between true HGS (i.e., directly measured HGS) and bias-free predicted HGS. The neural basis of true HGS and ∆HGS was investigated by correlating them with regional gray matter volume (GMV). Statistical analyses were performed to test their sensitivity to longitudinal changes in stroke and major depressive disorder (MDD) patients compared to matched healthy controls (HC). HGS could be accurately predicted using anthropometric and demographic features with high accuracy using linear support vector machine (SVM). Compared to true HGS, ∆HGS showed high reassessment reliability and stronger, widespread associations with GMV, especially in motor-related regions. Longitudinal analysis revealed that neither HGS nor ∆HGS effectively differentiated patients from matched HC at post time-point. The proposed ∆HGS score exhibited stronger correlations with GMV compared to true HGS, suggesting it better represents the relationship between muscle strength and brain structure. While not effective in differentiating patients from HC at post time-point, the increase in ∆HGS from pre to post time-point in patient cohorts may indicate improved utility for monitoring disease progression, treatment efficacy, or rehabilitation effects, warranting further longitudinal validation.

The human thalamus is composed of multiple nuclei that differ in structure and function. From early development onwards, these nuclei form reciprocal, nucleus-specific connections with the cerebral cortex, contributing to sensory and cognitive processing. In childhood and adolescence, a key period of neurocognitive development, these connections undergo widespread refinement, yet how developmental trajectories of thalamocortical connections vary across nuclei remains unknown. Here, we leveraged the Human Connectome Project in Development dataset (HCP-D, N = 604, age range 8-21) and segmented 10 thalamic nuclei using a segmentation approach optimized for intrathalamic contrast. Applying probabilistic tractography, we reconstructed nucleus-specific thalamocortical connections and charted their maturational profiles based on changes in fractional anisotropy (FA) using generalized additive models. We found FA to increase in thalamocortical connections, with nucleus-specific variation in temporal profiles and magnitude of age effects. Connections of core-cell-rich, sensory-projecting nuclei, such as the lateral geniculate nucleus, showed earlier maturational plateaus, whereas matrix-cell-rich, association-projecting nuclei, such as ventral anterior nucleus, showed more sustained maturation. This links maturational heterochronicity to thalamic organization of cell distribution and connectivity embedding. In parallel, functional thalamocortical connectivity decreased with age, with FA and functional connectivity age effects coupled in nucleus-connections showing prolonged maturation. Finally, concordant age effects in connectivity and nucleus volumes suggest that intra-nucleus remodeling may support refinement of structural connections while reducing thalamocortical functional synchrony. Together, our work reveals that thalamocortical maturation is anchored in the developmental and organizational heterogeneity of thalamic nuclei, offering a framework for understanding how diverse thalamic nuclei contribute to neurocognitive development.

Machine learning is accelerating biomedical research. Cross-validation is widely used to compare predictive performance — not only to benchmark algorithms, but also to inform scientific applications, such as ranking biomarkers. However, prediction performance estimates across cross-validation folds are not independent. Standard tests for comparing prediction performance (e.g., paired t-test) assume independence and can therefore inflate false positive rates. In a PRISMA-guided meta-analysis of 210 studies (impact factor ≥15, 1 June 2020 – 1 June 2025), we find that 97% ignored fold dependence when comparing prediction performance. This problem is ubiquitous across scientific fields and unaffected by impact factor, rigor-promoting policies, or open science practices. Simulations across 420 scenarios spanning four diverse datasets show that ignoring fold dependence leads to invalid false positive control in most settings. Repeated cross-validation further compounds this problem, with false positive rates rising toward 100% as the number of repetitions grows. Existing fold-dependence-aware tests rely on strong assumptions because the variance of fold-level statistics and the between-fold correlation cannot be disentangled under standard cross-validation. We therefore propose the SHARP (Split-HAlf RePeated) test, a simple modification to standard cross-validation that enables direct estimation of variance and correlation. Benchmarked against 12 tests, SHARP provides the best overall balance of false-positive control, statistical power, and confidence-interval calibration across simulation schemes. We conclude by providing best practices and reporting guidelines for valid model comparison inference in biomedical machine learning and beyond.

Promoting brain health is vital for well-being and reducing healthcare burdens. Brain health as measured with the Brain Age Gap (BAG) - the difference between chronological and predicted brain age- relates to many factors. However, a holistic view, integrating the range of factors an individual brain is exposed to, is missing for understanding how the exposome shapes brain health. After computing BAG as an indicator of grey matter (GM) health, we predicted it using machine learning based on 261 exposome variables (spanning biomedical, environmental, lifestyle, socio-affective, and early life domains) in UK Biobank participants. Exposome data can predict GM health with factors pertaining to cardiovascular and bone health, along with alcohol and smoking, nutrition and diabetes showing greater contribution to the prediction. In such domains, life period and duration of exposure appeared crucial. These findings call for early prevention in cardiovascular and metabolic health to promote life-long brain health.