A clustering-based method for estimating pennation angle from B-mode ultrasound images

Area of Science:

• Biomechanics
• Medical Imaging
• Robotics

Background:

• B-mode ultrasound (US) noninvasively measures skeletal muscle architecture, crucial for human intent in rehabilitation/assistive devices.
• Manual analysis of US images for muscle features is subjective, time-consuming, and labor-intensive.
• Accurate muscle feature extraction is vital for closed-loop human-robotic interaction control.

Purpose of the Study:

• To propose a novel clustering-based detection method for automatic identification of muscle fascicles and aponeurosis from B-mode US images.
• To compute the pennation angle accurately by mimicking a human expert's analysis.
• To evaluate the method's robustness and accuracy on low-frequency image streams.

Main Methods:

• A clustering-based detection algorithm assuming tubular muscle fiber characteristics was developed.
• The method identifies fascicle and aponeurosis to calculate the pennation angle.
• Performance was benchmarked against UltraTrack and ImageJ on a dataset with a 20 Hz frame frequency.

Main Results:

• The proposed clustering-based method demonstrated higher accuracy compared to UltraTrack and ImageJ.
• The algorithm's accuracy was comparable to that of a trained human expert.
• The method proved robust for low-frequency ultrasound image streams.

Conclusions:

• The developed method shows significant potential for automatic muscle fascicle orientation detection.
• This technique can facilitate biomechanics modeling, rehabilitation robot control design, and neuromuscular disease diagnosis.
• The approach is suitable for applications utilizing low-frequency ultrasound data streams.