Objective Assessment of Ideomotor Limb Apraxia Using Markerless Computer Vision and Machine Learning: A Proposed Framework and Validation Design.

Ideomotor limb apraxia is a higher-order disorder of learned, skilled movement that cannot be adequately explained by elementary weakness, sensory loss, ataxia, extrapyramidal dysfunction, impaired comprehension, or lack of cooperation. It is common after left hemisphere stroke and occurs in several neurodegenerative diseases. Quantitative kinematic study of apraxic movement is well established in the research literature, and standardized clinical instruments such as the Test of Upper Limb Apraxia and its bedside screen have improved the reliability of the examination. What remains missing is a bridge between these two bodies of work: an approach that uses only ordinary video, requires no laboratory motion capture, and maps recovered movement features onto a clinically interpretable profile of apraxic errors for routine bedside or remote use. This report advances the hypothesis that markerless computer vision, combined with machine learning trained against blinded expert consensus, can serve as an objective measurement adjunct for ideomotor limb apraxia: quantifying selected spatial and temporal features of gesture production, profiling error dimensions at the level of individual trials, and, in a more limited way, supporting detection and grading. The hypothesis is deliberately framed as a measurement adjunct rather than an autonomous diagnostic system. This report sets out the clinical rationale, the narrowed novelty claim, the proposed framework and its two analytical levels, the outcomes that would be predicted if the hypothesis holds, a validation design whose central feature is a control group of neurologically affected patients without apraxia, and the substantial limitations that bound the proposal. The framework specifies, as prespecified comparisons rather than as assumptions, alternative pose estimation families including transformer-based estimators, optional inertial augmentation with filter-based video inertial fusion, and alternative classifier architectures. Each is presented as a question the validation study should answer. This report presents a theoretical framework and a validation design only. No empirical patient data were collected, no patients were recruited, and no results are reported.