Novel computational protocol to support transfemoral prosthetic alignment procedure using machine learning...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Healthcare

Background:

• Prosthetic alignment is crucial for acceptable gait and preventing long-term disease.
• Current alignment assessment is subjective and relies heavily on prosthetist experience.
• Machine learning offers a potential solution to assist in optimal prosthetic alignment.

Purpose of the Study:

• To introduce a novel computational protocol for prosthetic alignment using machine learning.
• To aid prosthetists in accurately assessing and achieving optimal prosthetic alignment.
• To reduce subjectivity and improve consistency in prosthetic alignment procedures.

Main Methods:

• Trained machine learning models (SVM and neural network) on ground reaction force data from 16 transfemoral amputees with various alignments.
• Models predicted alignment condition, magnitude, and angle for correction.
• Validated the protocol with junior and senior prosthetists on two amputees.

Main Results:

• Support vector machine model achieved 92.6% accuracy in detecting nominal alignment.
• Neural network recovered correction angles with 94.11% accuracy and a 0.51° fitting error.
• Computational models and prosthetists showed agreement in alignment assessment, with high gait quality satisfaction scores (8/10 and 9.6/10).

Conclusions:

• The developed computational protocol effectively assists prosthetists in prosthetic alignment.
• This tool can decrease gait deviations and musculoskeletal diseases linked to misalignment.
• Improved alignment accuracy and consistency enhance amputee-prosthesis adherence and overall gait quality.