Digital Twins in Neuroscience: A Narrative Review of Technical Challenges for Child Neurology

• Digital twin (DT): a dynamically updated, computer-based replica of a real patient or organ that uses live data to predict and inform clinical decisions

• Proto-twin: an early-stage patient-specific computer model that simulates biology but doesn't yet meet the full criteria of a true digital twin (e.g., lacks continuous real-time updating)

• Digital biomarker: a digitally and objectively measured sign of a biological process (e.g., from imaging or EEG) used to track diagnosis, prognosis, or treatment response

• Verification, validation, and uncertainty quantification (VVUQ): a formal process for checking that a computational model is accurate, reliable, and its confidence limits are known

• Machine learning vs. deep learning: two types of AI: machine learning uses simpler statistical algorithms trained faster on less data; deep learning uses multi-layered neural networks needing more data but able to handle more complex prediction tasks

• Multiscale modelling: combining data and mechanisms from different biological levels (molecules, cells, whole organs) into one model

• EEG (electroencephalography): a test that records electrical activity in the brain

• Neurodevelopmental outcome: how a child's brain and nervous system function/develop over time (e.g., motor, cognitive skills)

• Interoperability: the ability of different computer systems or datasets to exchange and use information smoothly

1. Map the conceptual and technological development of digital twins (DTs) in neuroscience and pediatric healthcare

2. Examine representative "proto-twin" platforms (The Virtual Brain, The Virtual Child, etc.)

3. Critically evaluate the only two published child-neurology applications explicitly framed as digital twins (the infant microbiome DT and the neuromotor DT for preterm infants) assessing how close current technology comes to genuine, clinically deployable digital twins versus simpler patient-specific simulations.

The authors conducted a structured literature search (PubMed, Scopus, IEEE Xplore, Web of Science; January 2010–July 2026) and found that almost all neurological "digital twins" are actually proto-twins — patient-specific simulations that lack the continuous, bidirectional real-time updating and formal validation (VVUQ) that a true clinical-grade digital twin requires.

Of six neurological DTs identified in a prior scoping review, only one (UISS-MS, a multiple sclerosis immune-system simulator) had undergone formal VVUQ evaluation, and even that lacked prospective validation.

Representative platforms reviewed include The Virtual Brain (TVB) and The Virtual Epileptic Patient, which convert MRI data into 3D whole-brain simulations for research and presurgical epilepsy planning respectively, and

The Virtual Child, a cell-level model of pediatric cancer used for virtual clinical trials (conceptually similar to a DT but not formally defined as one). In adults, digital biomarkers show more clinical maturity — for example, an EEG-derived Alzheimer's disease model improved classification accuracy and predicted cognitive decline more accurately than conventional EEG biomarkers, though the authors caution these results still need external validation. Pediatric evidence is markedly limited, consisting of only two proof-of-concept applications:

(1) an infant microbiome digital twin (using gut bacteria profiles from 88 preterm infants across two U.S. sites) that models age-dependent bacterial interactions linked to neurodevelopmental outcomes, and

(2) a neuromotor digital twin for longitudinally monitoring preterm infants' motor and cognitive development using standardized clinical assessment data. Both are framed as promising but preliminary — observational and correlational rather than causally validated, and neither has undergone prospective multicenter validation.

The authors argue that child-neurology digital twins remain "in their infancy" and stress several structural gaps: no dedicated international guidelines currently exist for evaluating, systematically reviewing, or conducting meta-analyses of DT research (unlike related fields such as AI-based diagnostics, which have TRIPOD+AI, STARD-AI, CONSORT-AI, and FUTURE-AI reporting standards).

They draw useful methodological parallels from adjacent fields — perinatal hemodynamic modelling, cardiovascular DT research, and radiomics reproducibility frameworks (like METRICS and CLEAR) — as templates for how consensus-based reporting and validation standards could accelerate clinical readiness for neurological DTs.

For future progress, the authors recommend: developmentally aware longitudinal datasets that account for the rapidly changing anatomy and physiology of the developing brain; interoperable data standards (to allow different centers' data and models to work together); external and multicenter validation; transparent uncertainty reporting; and prospective implementation studies that demonstrate real clinical decision impact, not just retrospective or proof-of-concept accuracy.

They highlight childhood epilepsy — particularly severe epileptic encephalopathies like Lennox-Gastaut syndrome — as a priority area, given its high burden of illness and clear genetic underpinnings suitable for personalized modelling.

The authors close by framing digital twins as clinician-supervised decision-support tools, not autonomous replacements for clinical judgment, and raise open questions about minimum data requirements, how to adapt validation frameworks to a rapidly developing pediatric brain, how to prevent demographic/socioeconomic bias in federated pediatric datasets, and how consent and data governance should evolve as a child grows and their digital twin accumulates lifelong data.