200 Section - Development of an Automated Functional Prediction Rule for the Fugl-Meyer Score in Subacute and Chronic Stroke Patients

Wednesday, May 20, 2026

6:00 PM - 7:30 PM PT

Presenting Author/Autor expositor(s)

CA

César D. Amador Aburto, MD

PHYSICIAN SPECIALIST IN REHABILITATION
INSTITUTO NACIONAL DE REHABILITACION LUIS GUILLERMO IBARRA IBARRA
TLALPAN, Distrito Federal, Mexico

Co-Author/Coautor(s)

Alberto Isaac Pérez Sanpablo, PhD (he/him/his)

Researcher in science
INSTITUTO NACIONAL DE REHABILITACION LUIS GUILLERMO IBARRA IBARRA
Tlalpan, Distrito Federal, Mexico

Corresponding Author/Autor a quien dirigir la corr(s)

JQ

Jimena Quinzaños Fresnedo, PhD

PHYSICIAN SPECIALIST IN REHABILITATION
INSTITUTO NACIONAL DE REHABILITACION LUIS GUILLERMO IBARRA IBARRA
Tlalpan, Distrito Federal, Mexico

Objectives :

Objectives: To develop an automated functional prediction rule, based on artificial intelligence, capable of estimating the Fugl-Meyer Assessment (FMA) score—Total, Upper Extremity, and Lower Extremity domains—using routinely collected clinical and sociodemographic variables in medical practice.

Design:

Methods: An observational, longitudinal, retrospective study was conducted including 299 stroke patients treated at the Neurological Rehabilitation Department of the INR. Functional clinical data were collected, and predictive models (Random Forest, LASSO, and linear Support Vector Regression) were trained with cross-validation, using R², RMSE, and MAE as performance metrics.

Results:

Results: FMA scores showed strong correlations with functional indices: Barthel Index (ρ = +0.599), Upper Limb Motricity Index (ρ = +0.739), and Lower Limb Motricity Index (ρ = +0.692). Finger extension emerged as the most significant dichotomous clinical predictor. Random Forest applied to the Upper Extremity FMA achieved the best performance (R² = 0.891), followed by LASSO for Lower Extremity FMA (R² = 0.847), and linear SVR for Total FMA (R² = 0.719), all with clinically acceptable error ranges.

Conclusion:

Conclusions: The Motricity Index and Barthel Index effectively predict FMA scores during functional recovery after stroke. Finger extension, facial paresis, left-hemisphere lesions, and PACI territory involvement influenced prognosis. SVR, LASSO, and especially Random Forest models demonstrated high accuracy, proving useful for estimating recovery trajectories, guiding clinical decision-making, and optimizing rehabilitation.