Machine Learning Model for Predicting Walking Ability in Lower Limb Amputees.

Area of Science:

• Rehabilitation Medicine
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Rising incidence of lower limb loss (LLL) necessitates accurate assessment of walking potential.
• Informed prosthetic prescription and rehabilitation planning are critical for individuals with LLL.
• Predictive models can support clinical decision-making for LLL care.

Purpose of the Study:

• To develop a machine learning model for predicting walking ability in individuals with LLL.
• To identify key factors influencing walking potential and functional outcomes.
• To assist rehabilitation teams in prosthesis selection and patient counseling.

Main Methods:

• Prospective cross-sectional study of 104 participants with LLL.
• Data collection included demographics, physical, psychological, and social factors.
• Support Vector Machines (SVM) used to build predictive models for K-level, Timed Up and Go Test (TUG), and Two-Minute Walking Test (TMWT).

Main Results:

• Eight significant predictors identified for K-level, TUG, and TMWT: balance, BMI, age, depression, amputation level, and muscle strength.
• The Multidimensional Scale of Perceived Social Support (MSPSS) was an additional predictor for K-level.
• SVM models demonstrated high accuracy in predicting functional outcomes.

Conclusions:

• Machine learning, specifically SVM, can accurately predict functional ambulation outcomes in individuals with LLL.
• The developed predictive model can guide clinical practice and inform patients about their mobility potential.
• Integration of these assessments into routine care can optimize rehabilitation strategies and prosthesis fitting.