Motor Cognitive Reserve 2.0: Equitable AI-Supported Stress Testing of Motor–Cognitive Reserve for Early Neurodegeneration

Background

Neurodegenerative diseases such as Alzheimer’s disease (AD) and Parkinson’s disease (PD) may be biologically detectable years before overt symptoms emerge. However, biomarkers alone do not explain why some individuals develop early motor or cognitive decline while others remain functionally resilient. This variation is commonly described as reserve: the capacity to maintain function despite accumulating pathology. Existing measures, such as education, IQ, or brain volume, are static proxies and may not capture an individual’s changing capacity to cope with increasing cognitive and motor demands.

Neurodegenerative diseases are also shaped by health inequalities. Opportunities for early recognition and participation in prevention studies may differ by socioeconomic circumstances, education, ethnicity and more. Conventional proxies for cognitive reserve  may partly reflect unequal social opportunities rather than underlying resilience. A precision-medicine approach must therefore establish whether motor–cognitive reserve measures are valid and usable across diverse groups.

This PhD will will develop and evaluate interpretable and equitable machine-learning methods for personalised assessment of motor–cognitive reserve in early neurodegeneration. The project will be based on a collaborative dataset that includes patients using  virtual-reality treadmill environment with motion capture and wearable inertial sensors. Data includes a personalised MCR score from performance under increasing demand, including cognitive accuracy, reaction time, obstacle-negotiation success, and gait variability. Preliminary evidence suggests that the MCR index distinguishes healthy participants from people with neurological diagnoses more effectively than conventional reserve proxies (Koslovsky et al., 2023; Koslovsky et al., 2025).

The PhD project will extend the global MCR score by identifying cognitive and motor reserve profiles associated with AD- and PD-related biomarker pathways, relating these profiles to resting-state functional MRI connectivity, and contributing to a real time version of the assessment. It will use historical MCR data and a prospective longitudinal cohort comprising 40 individuals with alpha-synuclein (PD) or pTau/amyloid (AD) biomarker profiles, alongside biomarker-negative healthy controls.

Aims

The PhD will develop and evaluate interpretable machine-learning methods for personalised assessment of motor–cognitive reserve in early neurodegeneration.

1. Develop domain-specific MCR models. The student will model motor, cognitive, and combined motor–cognitive performance using task success, reaction times, dual-task costs, and gait features derived from inertial sensors. Models will predict MCR scores from these features.
2. Model change and heterogeneity. Using baseline and follow-up assessments, the student will investigate whether MCR indices predict change in motor, cognitive, and clinical outcomes. The work will compare transparent statistical and regularised models with non-linear methods, prioritising calibration, uncertainty, subgroup performance, and interpretability.
3. Integrate behavioural and neuroimaging evidence. The project will examine associations between MCR indices and resting-state functional connectivity, using network-based measures and data-driven approaches—including independent component analysis—to identify neural correlates of reserve.
4. Contribute to ADAPT-MCR. The PhD will support development of an adaptive assessment algorithm that selects informative next tasks, estimates uncertainty, and applies safe stopping criteria. The aim is to reduce testing burden while retaining agreement with the full MCR protocol and clinically meaningful differentiation of motor and cognitive pathways.
5. Evaluate equity risks using simulation and sensitivity analysis. The student will examine whether MCR measurement properties, model calibration, uncertainty, error rates differ across demographic, socioeconomic, and clinically relevant groups represented in the data. Sensitivity analyses, supplemented by simulations where subgroup data are limited, will assess the effects of differential missingness, measurement error, and population shifts.

Training Outcomes

The student will receive interdisciplinary training in artificial intelligence, digital health, neuroscience, and clinical neurodegeneration. They will gain skills in multimodal and longitudinal machine learning, wearable-sensor and behavioural-data analysis, clustering, latent-variable modelling,  and robust model evaluation.

Training will cover reproducible research, secure health-data management, responsible AI, fairness assessment, uncertainty communication, explainability methods such as SHAP, and clinically meaningful validation.

Working across the University of Edinburgh and the data suppliers (Tel Aviv University Medical Center), the student will collaborate with AI researchers, neurologists, imaging specialists, and clinical researchers. They will develop the skills to design trustworthy clinical decision-support tools that complement professional clinical judgement.