Developing Fine-Grained Actigraphies for Rheumatoid Arthritis Patients from a Single Accelerometer Using Machine...

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Wearable Technology

Background:

• Rheumatoid Arthritis (RA) significantly impacts patients' quality of life, increasing sedentary behavior and hindering daily activities.
• Objective monitoring beyond clinical exams is crucial for understanding RA's effect on patient mobility.
• Existing methods may not capture the fine-grained activity changes caused by RA.

Purpose of the Study:

• To develop and validate a method for generating detailed activity recognition (actigraphy) from accelerometer data in RA patients.
• To automatically tag accelerometer data, capturing the impact of RA on daily physical activities.
• To create objective, RA-specific mobility markers for use between clinical visits.

Main Methods:

• A processing methodology using machine learning and deep learning was developed to tag accelerometer data.
• Thirty subjects (10 RA patients, 20 controls) wore a tri-axial accelerometer at the L5 vertebra.
• The method handles unbalanced datasets, long activities (sitting, lying), short transitions (sit-to-stand), and includes confidence thresholding and logical filtering.

Main Results:

• The developed method achieved approximately 95% accuracy and an 81% F-score in activity prediction.
• The system successfully captured fine-grained activities and transitions relevant to RA patients.
• The methodology demonstrated robustness in handling data challenges like unbalanced datasets and complex activity sequences.

Conclusions:

• The proposed method effectively generates detailed actigraphies for RA patients using accelerometer data.
• This approach offers objective, RA-specific markers of patient mobility, aiding in disease management.
• The technology has the potential to enhance remote patient monitoring and clinical assessment of RA progression.