Optimizing Stroke Detection Using Evidential Networks and Uncertainty-Based Refinement.

Evaluating neurological impairments post-stroke is essential for assessing treatment efficacy and managing subsequent disabilities. Conventional clinical assessment methods depend largely on clinicians' visual and physical evaluations, resulting in coarse rating systems that frequently miss subtle impairments or improvements. Interactive robotic devices, like the Kinarm Exoskeleton system, are transforming the assessment of motor impairments by offering precise and objective movement measurements. In this study, we analyzed kinematic data from 337 stroke patients and 368 healthy controls performing three Kinarm tasks. Using deep learning methods, particularly an evidential network, we distinguished impaired participants from healthy controls while generating measures of prediction uncertainty. By retraining the network with the least uncertain samples and refining the test set by excluding the top 10% most uncertain samples, we improved the sensitivity of detecting subtle impairments in minimally impaired stroke patients (those scoring normal on the CMSA) from 0.55 to 0.75. We further extended the model to detect impairments associated with transient ischemic attack (TIA), resulting in an increased detection accuracy from 0.86 to 0.92. The model's ability to identify subtle motor deficits, even in TIA patients who show no observable symptoms on standard clinical exams, highlights its significant clinical utility. Detecting TIA is critical, as individuals who experience a TIA have a substantially higher risk of recurrent stroke. This work highlights the immense potential of integrating deep learning with uncertainty estimation to enhance the detection of stroke-related impairments, potentially paving the way for personalized post-stroke rehabilitation.