Related Experiment Video
Updated: Aug 6, 2026
Motor Imagery Brain-Computer Interface in Rehabilitation of Upper Limb Motor Dysfunction After Stroke
Published on: September 1, 2023
Predicting Motor FIM at Discharge for Recovery Stroke Patients in a Rehabilitation Ward Using an AI Predictive
Kazuyuki Morita1, Senshu Abe2,3, Keisuke Ono2
1Occupational Therapy Section, Rehabilitation Department, Tokachi Rehabilitation Center, Hokuto Social Medical Corporation.
Abstract:
Morita K, Abe S, Ono K, Takahashi R, Ara Y, Abe M, Shirasaka T. Predicting Motor FIM at Discharge for Recovery Stroke Patients in a Rehabilitation Ward Using an AI Predictive Analysis Tool. Jpn J Compr Rehabil Sci 2026; 17: 40-45.
Objective:
This study aimed to create a prediction model for motor Functional Independence Measure (FIM) at discharge using an artificial intelligence (AI) predictive analysis tool (Prediction One, Sony Network Communications Inc.) and to clarify its accuracy.
Methods:
This study included 457 stroke patients discharged between April 2020 and March 2022 as the training dataset, and 43 stroke patients discharged between April and October 2022 as the validation dataset. First, an AI prediction model was developed using the training dataset. Next, the prediction accuracy was evaluated using the validation dataset. As part of the accuracy validation, the residuals between the actual motor FIM at discharge and the predicted values of the AI and therapists, as well as the correlation coefficients and coefficients of determination for each prediction result were calculated and statistically compared.
Results:
The coefficient of determination for the prediction model was 0.799. The residuals from the actual values were 5.43 ± 5.93 points for AI predictions and 6.42 ± 5.52 points for therapist predictions. The correlation coefficients and R2 values for the prediction results were 0.939 and 0.881 for AI predictions, and 0.943 and 0.889 for therapist predictions, respectively. No significant difference was observed between the two correlation coefficients (z = -0.295, p = 0.768). The residual distributions indicated that the therapist predictions tended to overestimate the actual values.
Conclusion:
The model developed in this study demonstrated an accuracy no less than that reported in previous studies. Furthermore, it achieved a predictive performance comparable to that of the therapists. This model could potentially predict motor FIM at discharge with consistent accuracy, regardless of the years of experience. In the future, we plan to expand the range of target conditions and assess their external validity.
