Muscle Type and Gender Recognition Utilising High-Level Textural Representation in Musculoskeletal Ultrasonography.

Area of Science:

• Medical Imaging
• Computer-Aided Diagnosis
• Machine Learning

Background:

• Human assistive technology and computer-aided diagnosis are advancing in medical imaging.
• Current gray-scale image analysis for musculoskeletal conditions is limited by image acquisition systems.

Purpose of the Study:

• To investigate machine learning feature extraction techniques for musculoskeletal ultrasonography.
• To assess the performance of these techniques in identifying human characteristics from muscle images.

Main Methods:

• A new dataset of 74 healthy patients' musculoskeletal ultrasound images was created.
• Two classification tasks were performed: gender and muscle-type recognition.
• A novel computer vision method was used for high-level feature representation extraction.

Main Results:

• The method achieved an 87.07% classification rate for muscle-type recognition.
• Gender recognition performance surpassed state-of-the-art textural representations in most muscle sections.

Conclusions:

• Advanced feature extraction techniques show significant potential for improving musculoskeletal ultrasound analysis.
• This approach offers a more robust method for distinguishing between healthy and pathologic conditions.