White Matter Structural Biomarkers Derived from Diffusion Tensor Imaging to Estimate Upper-Extremity Motor Function in Stroke Using Machine Learning.

Upper extremity (UE) motor function assessment of stroke patients is essential for diagnosis and rehabilitation planning. Clinical evaluations, such as the Fugl-Meyer Assessment for the Upper Extremity (FMA-UE) and the Action Research Arm Test (ARAT), are commonly used to measure stroke patients’ UE motor capabilities. On the other hand, biomarkers derived from white matter integrity features have shown associations with UE motor function and could be coupled with machine learning algorithms to estimate FMA-UE and ARAT within stroke population. For this reason, in this work, the Feature-ranked Self-growing Forest (FSF) algorithm was used to estimate the UE sensorimotor function and functional performance of stroke patients, by estimating FMA-UE and ARAT scores from fractional anisotropy ratios from homologous regions of interest within the ipsilesional and contralesional hemispheres derived from diffusion tensor imaging data. The obtained estimations for the FMA-UE and the ARAT had significant associations (FMA-UE: rs = 0.79, p < 0.001. ARAT: rs = 0.71, p < 0.001) with actual assessments’ scores, and there were not statistically significant differences between estimated and actual assessments. Moreover, the most relevant white matter regions for performing the estimations of both FMA-UE and ARAT were those associated in the literature with patients’ performance during the execution of UE motor tasks and the severity of spasticity, such as the corticospinal tract at the brainstem level, the superior corona radiata, and the external capsule. Therefore, this study shows that there are nonlinear associations of white matter integrity features from homologous regions of interest within the ipsilesional and contralesional hemispheres, with UE function that could be used as an auxiliary objective tool for stroke patients’ clinical assessment.