Research Agenda · 2026–2028
Our research programme is selected against three criteria: clinically or scientifically urgent, resistant to solution despite significant prior effort, and tractable using formal and computational methods.
Five foundational gaps
These are not incremental research questions. They are foundational gaps that have resisted decades of effort.
AI clinical decision tools are deployed across health systems with no mathematical proof that they cannot recommend harmful treatment sequences. The MHRA 2026 AI device framework requires safety cases. No accepted formal methodology exists.
Neurosurgeons still drill skulls to measure brain pressure. Every non-invasive approach — transcranial Doppler, optic nerve sheath, tympanometry — fails calibration. The 95% limits of agreement remain ±7–15 mmHg, clinically unacceptable.
Sepsis kills 11 million people annually. Existing AI tools detect sepsis after clinical recognition — when the patient is already deteriorating. The pre-clinical trajectory has never been formally modelled. Point-in-time scores cannot reason about trajectory dynamics.
The wearables industry has invested billions without asking a foundational question: what physiological information is theoretically recoverable from surface measurements, given the physics of signal propagation through biological tissue?
Hospitals are buying correlation engines. Clinical AI predicts but cannot answer counterfactual questions: what would happen if we intervened now versus in four hours? The gap between predictive and causal clinical AI has not been bridged by any deployed system.
Twelve research topics
Topics are tiered by time to first publishable output. Each is grounded in Vitacore's formal-methods identity.
Auscultation accuracy is clinician-dependent and inconsistent. Existing AI classifiers lack formal noise robustness guarantees and operate only in ideal acoustic conditions.
Funding: NIHR Digital Health · NHS AI regulation pathway
Speech biomarkers for neurological and respiratory conditions lack formal signal chain specification. No regulatory framework defines minimum signal quality requirements.
Funding: MHRA 2026 AI device framework · NIHR digital biomarker streams
Sepsis early-warning systems are point-in-time statistical models. No formal model of pre-clinical physiological trajectory exists. Alarm-threshold verification is ad hoc.
Funding: NIHR 2026/27 strategic funding · Innovate UK AI in healthcare
There is no foundational analysis of what physiological information is theoretically recoverable from surface measurements. Sensing research proceeds without knowing fundamental limits.
Funding: EPSRC theoretical foundations · basis for all sensing grants
Non-invasive intracranial pressure has remained unsolved for thirty years. The missing piece is a principled signal-fusion framework grounded in the physics of propagation through brain tissue.
Funding: NIHR i4i Product Development · Innovate UK · DASA
Bone stress fractures and joint degradation produce acoustic emission signatures identical to structural micro-fracture in civil engineering. Nobody has formally bridged these disciplines.
Funding: EPSRC materials and healthcare crossover · Innovate UK
Geophysics uses sophisticated deconvolution and source-separation algorithms for micro-seismic events. Neurological tremor is the same signal-processing problem at human scale — a bridge never formally made.
Funding: UKRI/EPSRC cross-disciplinary · Parkinson's UK · MRC
WiFi and RF signals already permeate hospitals. Changes in propagation caused by breathing and heartbeat are detectable without worn devices. No formally specified, regulatory-compliant acquisition architecture exists.
Funding: NIHR digital health · NHS monitoring for resistant cohorts
Pulmonary oedema and fluid density gradients in ocean layers are physically identical acoustic propagation problems. Pulmonologists have not applied oceanographic methods.
Funding: NIHR respiratory · BHF · EPSRC cross-disciplinary
Phase transitions in engineering materials — steel fatigue, composite delamination, concrete micro-cracking — are described statistically. No formal automata model of phase-transition dynamics enables verified prediction of failure reachability.
Funding: EPSRC materials · Innovate UK infrastructure · defence and aerospace
Every hospital system purchasing AI is buying correlation. No deployed system can reason causally about counterfactual clinical questions. This is the most important limitation of clinical AI.
Funding: NIHR · Wellcome Trust digital health · UKRI AI for health
Non-invasive neuromodulation devices are being developed without formal specification of closed-loop control behaviour. The MHRA 2026 framework will require it.
Funding: NIHR i4i · Innovate UK Biomedical Catalyst
The hidden engines
Our methodological toolkit is domain-agnostic by construction. The same mathematics that characterises one signal characterises another in a different field.
Persistent homology of physiological signals.
Detects structural changes in waveforms invisible to amplitude-based methods. Applied across sepsis trajectory, ICP morphology, tremor analysis.
Stack: giotto-tda · ripser
Parameter-free complexity from compression.
Measures signal complexity collapse — a proven early marker of physiological deterioration. Approximates Kolmogorov complexity practically.
Stack: zlib · standard library
Signal processing on irregular network structures.
Models organ-system coupling, brain connectivity, RF propagation networks. Spectral and filtering operations on graph-structured signals.
Stack: PyGSP · NetworkX
Formal specification of communicating processes.
Verifies that clinical pathways and closed-loop device protocols cannot deadlock or reach unsafe states. Direct application to MHRA safety cases.
Stack: FDR4 · CSP-M
Formal models of timed system dynamics.
Specifies physiological state transitions with verified timing properties. Reachability analysis for early-warning alarm verification.
Stack: UPPAAL · PRISM
Counterfactual reasoning machinery.
Provides the formal foundations for digital twins that can answer counterfactual clinical questions — what would happen if we intervened now versus later?
Stack: DoWhy · CausalNex
Shannon capacity and mutual information.
Bounds the theoretical recoverability of physiological state from surface measurements. Provides the channel capacity of clinical assays (e.g. ctDNA detection).
Stack: Information-theoretic derivations
Multipath cancellation, channel estimation.
Direct transfer from telecommunications. Applied to acoustic and electromagnetic propagation through biological tissue.
Stack: SciPy signal · custom