سامانه همایش های دانشگاه آزاد اسلامی واحد نجف آباد

صفحه اصلی / اولین همایش بین المللی تحول خدمات سلامت: تلفیق مراقبت، درمان، بازتوانی و امیدبخشی

Computer Vision-Based Goniometer: Measuring Passive Range of Motion in the Knee Using a Markerless Deep Learning Approach

نویسندگان :

Sina Alizadeh Tabrizi¹

1- دانشجوی دکتری، گروه مهندسی کامپیوتر، دانشگاه آزاد اسلامی، واحد ارومیه

کلمات کلیدی :

Knee range of motion،Goniometry،Computer vision،Deep learning،Pose estimation،Inter-clinician variability،Rehabilitation

چکیده :

Background: Manual goniometry, the gold standard for knee range of motion (ROM) measurement, has high inter-rater variability, making treatment tracking challenging. To validate a computer vision-based goniometer that uses deep learning to measure passive knee ROM without markers. Special emphasis was placed on minimizing inter-rater variability Method: Forty-five subjects (90 knees) underwent passive flexion and extension ROM measurements by two raters using manual goniometry and computer vision. The computer vision system used MediaPipe Pose for keypoint identification (hip, knee, ankle), Kalman filtering, and automated peak detection. Reliability was evaluated using intraclass correlation coefficients (ICC), and validity using Bland-Altman analysis. Inter-rater variability was compared between the two methods. Finding: The computer vision system showed high intra-rater (ICC > 0.93) and inter-rater reliability (ICC > 0.91), outperforming manual goniometry (inter-rater ICC = 0.71-0.76). Bland-Altman plots revealed high agreement, with mean biases of -1.2° (flexion) and 0.8° (extension). Crucially, the system reduced inter-clinician variability by 55-56%, with mean absolute differences between raters decreasing from 7.1° to 3.2° for flexion (p < 0.001) and 4.8° to 2.1° for extension (p < 0.001). Conclusion: The markerless deep learning-based goniometer achieves clinically acceptable accuracy while substantially reducing inter-clinician variability compared to manual goniometry, potentially enabling standardized ROM measurement in clinical rehabilitation settings.