Digital Balance Biomarkers and Interpretable Evaluation of Disease Severity in Spinocerebellar Ataxia Type 3

The clinical assessment of spinocerebellar ataxia type 3 (SCA3) is hindered by subjective factors and inter-rater variability. This study utilizes a previously validated wearable plantar pressure insole system established in our prior work to build a multiscenario digital balance detection framework, and assesses its efficacy in differentiating SCA3 patients from healthy individuals and in evaluating disease severity. The study involved 74 SCA3 patients and 45 healthy controls, who were equipped with custom-developed plantar force sensors to perform standing tasks with eyes open (EO) and eyes closed (EC). Center of pressure (COP) variables were analyzed to determine their correlation with clinical characteristics and their discriminative capability. An XGBoost model was employed to classify disease severity, with SHAP analysis providing enhanced interpretability. COP variables derived from EC tasks demonstrated superior performance in terms of area under the curve, accuracy, sensitivity, and specificity compared to those from EO tasks. Significant correlations were identified between COP variables, such as the eyes closed, left side sway velocity (EC-L velocity) SD, and clinical scores (r > 0.4). The model achieved an accuracy of 87% in classifying disease severity, with SHAP analysis highlighting key variables and interactions. COP variables offer a robust, multidimensional measure of motor dysfunction.