Automatic Measurement of Postural Abnormalities With a Pose Estimation Algorithm in Parkinson's Disease

Area of Science:

• Biomedical Engineering
• Neurology
• Computer Vision

Background:

• Parkinson's disease (PD) is a neurodegenerative disorder characterized by motor symptoms, including significant postural abnormalities.
• Accurate and objective assessment of these postural changes is crucial for diagnosis, monitoring disease progression, and evaluating treatment efficacy.
• Current methods for evaluating posture in PD patients often rely on subjective clinical assessments or time-consuming manual measurements.

Purpose of the Study:

• To develop and validate an automated, objective tool for quantifying postural abnormalities in Parkinson's disease patients.
• To utilize deep learning-based pose estimation for precise measurement of key postural parameters.
• To establish the reliability of automated measurements against conventional manual methods.

Main Methods:

• A deep learning-based pose-estimation algorithm was applied to lateral photographs of 28 Parkinson's disease patients.
• The algorithm automatically calculated the anterior flexion angle (AFA) and dropped head angle (DHA).
• Automated measurements were rigorously validated against traditional manual labeling techniques.

Main Results:

• The automated measurements of DHA and AFA demonstrated excellent agreement with manual labeling.
• The intraclass correlation coefficient (ICC) for both angles exceeded 0.95, indicating high reliability.
• The mean bias between automated and manual measurements was minimal, equal to or less than 3 degrees.

Conclusions:

• The developed deep learning-based pose-estimation algorithm provides an objective and accurate method for assessing postural abnormalities in Parkinson's disease.
• This automated tool has the potential to enhance clinical evaluation and research in PD.
• The high agreement with manual methods supports the clinical utility of this novel approach.