Using Wearable Sensors and Machine Learning Models to Separate Functional Upper Extremity Use From Walking-Associated...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Healthcare

Background:

• Accurate measurement of upper extremity (UE) use is crucial for evaluating prosthetic device effectiveness and rehabilitation progress.
• Current methods often rely on subjective assessments or limited observational data, failing to capture real-world functional activity.
• Distinguishing purposeful UE movements from non-functional movements like walking is a significant challenge in objective UE use measurement.

Purpose of the Study:

• To evaluate the feasibility of using body-worn inertial sensor data and machine learning models to differentiate functional upper extremity (UE) use from walking-related movements.
• To develop and assess classification models capable of distinguishing productive prehensile and bimanual UE activity from extraneous movements.
• To improve the objective measurement of UE use in community settings for individuals with UE limb loss.

Main Methods:

• Comparison of machine learning classification models against a criterion standard of manually scored video data.
• Utilized inertial sensor data from the dominant wrist synchronized with video recordings of participants performing functional activities.
• Trained three classification models using data from upper extremity (UE) prosthesis users and controls, evaluating performance within-subject and across-subject.

Main Results:

• Machine learning models achieved high classification accuracy, correctly identifying functional UE movements versus walking in over 95% of test data across most trial types.
• The most effective model in the amputee/across-subject trial correctly classified 85% of test examples.
• Computationally lightweight models demonstrated reliability in distinguishing functional UE movements from walking-associated movements.

Conclusions:

• Body-worn inertial sensors combined with lightweight machine learning models can reliably differentiate functional prosthetic upper extremity (UE) use from walking.
• This approach shows significant promise for objectively quantifying real-world UE activity in individuals using prosthetic limbs.
• The findings suggest potential applications in clinical trials and for monitoring treatment response in various UE pathologies.