What the research found
Healthcare systems in the US face a substantial gap between demand for neurological expertise and specialist availability, particularly for Alzheimer's disease and related dementias. The authors propose that AI systems trained on large language models could help bridge this gap by automating and augmenting clinical workflows—integrating data from multiple sources (imaging, labs, clinical notes, cognitive assessments) and applying knowledge from specialist practice to non-specialist settings.
The paper outlines a six-stage implementation roadmap: standardised multimodal data collection; AI-assisted decision support for clinicians; workflow integration; rigorous validation and safety monitoring; iterative improvement based on real clinical feedback; and governance frameworks. The emphasis is on AI as a tool that enhances rather than replaces clinical judgment, with clinicians retaining responsibility for interpretation and decision-making.
Why it matters for you
If you're tracking biomarkers and using data to optimise your health, you're already doing what this roadmap advocates: assembling multiple data streams to build a complete picture. The bottleneck it identifies—clinicians lacking time or bandwidth to integrate complex information—applies equally to early cognitive screening, which often happens in primary care rather than neurology. Better AI-assisted integration of cognitive tests, blood biomarkers (tau, phospho-tau, amyloid), imaging findings and clinical history could catch earlier-stage cognitive decline in routine check-ups, when disease-modifying therapies are most likely to help.
For users monitoring longevity and neurological health, this matters because access to timely diagnosis currently varies by geography and specialist availability. If validated AI systems can reliably assist non-specialist clinicians in interpreting complex multimodal data, it becomes more feasible to identify candidates for early intervention (e.g., lecanemab, donanemab) outside tertiary centres. It also suggests that longitudinal biomarker tracking through apps like MyKine could increasingly feed into standardised clinical decision support—your data becomes actionable input rather than a standalone record.
Caveats
- Purely conceptual: This is a roadmap and vision paper, not a report of deployed or validated systems. No clinical outcomes or performance data are presented.
- Early-stage AI: Large language models in medicine are still being evaluated for accuracy and safety; real-world performance in dementia diagnosis remains unproven.
- Implementation risk: Integrating AI into clinical workflows requires regulatory approval, validation studies, and addressing liability and equity concerns—all outlined as future work, not completed.
- Specialist knowledge encoded unclearly: The paper assumes AI can reliably "learn" from practising specialists, but doesn't detail how that transfer occurs or how errors propagate.
- Scope limited to diagnostics: Therapeutic decision-making and long-term management are not addressed.