Table 2.

Challenges, possible solutions, limitations and hurdles for neuroimaging to support translational medicine in dementia

Challenge	Solutions	Limitations and hurdles
Reproducibility crisis and underused data	Open data in BIDS format, infrastructure like the Dementias Platform UK Portal, analysis code publication.	Very large file sizes, especially for neurophysiological data.
Requirement for large cohorts in clinical trials	Multisite cooperation through research networks that benefit all collaborators.	Funding climates that prioritise competitive over collaborative research.
Limited clinical, research and imaging resources to undertake deep and multimodal phenotyping	Standardised protocols. Collaborative grant funding. Cross-scanner harmonisation efforts like the UK7T partnership.	Differing research goals between centres. Locally optimal protocols may not be maximally transferrable.
Non-specific binding of PET ligands	Studies across comprehensive patient cohorts, multivariate pattern-analysis methods, neuropathological validation.	Large no of ligands from different vendors makes comprehensive characterisation challenging.
Slow translation to the clinic	Interdisciplinary work, collaboration with pharmaceutical companies in the design phase.	Inertia and a preference for insensitive but established measures in trial design and regulatory approval.
A proliferation of measures with unclear relative sensitivity and predictive value.	Head-to-head comparison studies in the same participants.	Studies are expensive, and rely on clinical tests that may be insensitive.
Large-scale clinical trials are hugely expensive, and not all mechanisms can be explored.	Small-n pharma coimaging studies to demonstrate proof of concept, motivating larger trials by rescuing functional biomarkers.	Rescuing imaging biomarkers and restoring neurotransmitter balance may be insufficient to provide clinical efficacy.
Individual variation within patient cohorts can mask real effects.	Multimodal neuroimaging to provide post hoc explanations of subgroup efficacy, leading to personalised medicine.	Prespecifying such analyses is often difficult, and false-positive associations mean insights become less trustworthy and generalisable.
Big data become increasingly difficult to analyse, and statistically significant findings can have small effect sizes or be driven by hidden bias.	Formal statistician involvement. Prespecified analyses. Artificial intelligence and machine learning techniques.	As algorithms become more complex, they can become less transparent and interpretable for patients and clinicians.

BIDS, brain imaging data structure; PET, positron emission tomography.