The investigators converted Luria’s classic fist–edge–palm paradigm into a self-administered smartphone assessment within the Neurological Functional Test Suite (NeuFun-TS). The objective was to capture granular subprocesses of dynamic praxis—specifically cognitive planning and motor execution—using touch and temporal data available from consumer devices.
A total of 296 participants completed the NeuFun-TS motor sequencing task. The sample comprised 34 healthy donors, 208 people with multiple sclerosis (MS), and 54 neurological control participants. Ethical approval for the study was provided by the Institutional Review Board of the National Institutes of Health.
From the smartphone recordings the authors derived seven digital biomarkers representing metrics across four domains: speed, execution, accuracy, and integrative throughput. These biomarkers were designed to quantify distinct aspects of motor sequencing performance and to enable decomposition into cognitive and motor subprocesses.
Six of the seven biomarkers significantly differentiated participant groups along an ordinal severity gradient (healthy donors < relapsing–remitting MS < progressive MS). The biomarker labeled Motor Sequencing Deficit produced the strongest group separation (correlation r = 0.73, p = 1.1 × 10⁻²¹), indicating large effect size for disease-related differences in motor sequencing on the smartphone task.
Kinematic analysis of the recorded movements allowed separation of cognitive planning latency from pure motor execution. Reported effect-size metrics indicate a small contribution from planning latency (ϵ2 = 0.011) relative to a much larger contribution from motor execution (ϵ2 = 0.716), evidencing that the task can dissociate subprocesses within dynamic praxis.
A speed–accuracy tradeoff analysis revealed distinct performance phenotypes among progressive MS subtypes. Secondary-progressive MS participants showed a predominantly slow-and-inaccurate profile (84%), whereas primary-progressive MS participants more often exhibited a slow-but-accurate, compensatory phenotype (19%). These distinctions suggest the smartphone task can capture heterogeneity of cognitive-motor strategies across MS phenotypes.
A concise two-biomarker composite trained on the NeuFun-TS features achieved strong diagnostic discrimination, with a concordance index of 0.868. The composite showed robust reliability on validation, with an intraclass correlation coefficient (ICC) = 0.77, supporting reproducibility of the selected digital measures.
The authors report that the smartphone-derived composite and individual biomarkers correlated with independent measures of central nervous system damage and clinical status. Correlation coefficients with MRI-assessed tissue destruction ranged from rho = 0.29 to 0.38. Correlations with clinician disability scales ranged rho = 0.44 to 0.52, and with cognitive performance scores ranged rho = 0.49 to 0.61. These associations link the digital motor sequencing measures to structural, clinical, and cognitive markers of disease burden.
The preprint states that all data produced will be available online in a GitHub repository upon journal acceptance (https://github.com/Bielekova-Lab/neufun-motor-sequencing). The work was reviewed and approved by the NIH Institutional Review Board. The authors declared no competing interests.
By isolating cognitive-planning and motor-execution components and by providing scalable, low-burden remote assessment, the NeuFun-TS smartphone motor sequencing task offers a framework for longitudinal monitoring of functional impairment in multiple sclerosis and other central nervous system disorders. The reported ability to discriminate disease severity, distinguish phenotypic performance patterns, and correlate with MRI and clinical measures supports its potential as a digital biomarker for research and clinical follow-up.
Details on specific algorithmic implementations, the exact definitions of the seven biomarkers, and full methodological parameters (for example, preprocessing steps and model training procedures) are provided in the source manuscript and supplementary material. The preprint indicates data and code availability at the linked GitHub repository for reproducibility and further inspection.