Keypoint detection and functional evaluation of human lower limbs based on YOLOv8 and HRNet

Area of Science:

• Computer Vision
• Artificial Intelligence
• Biomedical Engineering

Background:

• Human pose estimation is crucial for medical rehabilitation and motion analysis.
• Objective assessment of lower limb function in knee joint diseases requires advanced methods.
• Existing methods may lack accuracy and objectivity in functional assessments.

Purpose of the Study:

• To develop an AI-driven method for detecting and assessing lower limb keypoints and function.
• To integrate YOLOv8 and an enhanced HRNet for precise human pose and knee joint analysis.
• To evaluate the system's accuracy in measuring knee flexion-extension angles and sit-to-stand actions.

Main Methods:

• Utilized YOLOv8s for initial human region detection.
• Employed an improved HRNet-W32 with Gating Unit and Keypoint Attention Unit for detecting six lower limb keypoints (hip, knee, ankle).
• Calculated knee joint flexion-extension angles and recognized sitting-to-standing actions using a threshold state machine.

Main Results:

• The improved HRNet demonstrated a 3.9% increase in mAP@0.5 and 1.3% in PCK@0.3 compared to the original HRNet-W32.
• Achieved 97.84% average counting accuracy and 0.18 average absolute error for the 30-Second Sit-to-Stand Test.
• Demonstrated strong correlation between the algorithm's output and established clinical functional status indicators (KOOS, WOMAC, TUG).

Conclusions:

• The proposed integrated AI model offers an accurate and objective approach for assessing lower limb function in patients with knee diseases.
• The enhanced HRNet significantly improves keypoint detection performance.
• This method shows potential for clinical application in rehabilitation and osteoarthritis management.