Annotation-guided encoder-decoder network for bone extraction in ultrasound-assisted orthopedic surgery.

Area of Science:

• Medical Imaging
• Orthopedic Surgery
• Biomedical Engineering

Background:

• Fluoroscopy-based navigation in orthopedic surgery exposes patients and surgeons to significant ionizing radiation.
• Ultrasound-assisted surgery offers a radiation-free alternative with rich navigation data, but bone structure extraction from ultrasound images is challenging due to artifacts.
• Accurate bone structure extraction is crucial for effective ultrasound-assisted orthopedic navigation.

Purpose of the Study:

• To develop an effective method for extracting bone structures from ultrasound sequences for radiation-free orthopedic navigation.
• To address the challenges of image artifacts and probe pose variability in ultrasound-assisted orthopedic surgery.
• To improve the accuracy and speed of bone surface extraction for intraoperative guidance.

Main Methods:

• Proposed an annotation-guided encoder-decoder network (AGN) for bone structure extraction from ultrasound sequences.
• Incorporated a feature alignment module to handle variations in ultrasound frames caused by probe movement.
• Integrated annotation information from Siamese networks to enhance target extraction, mimicking interactive ultrasound analysis.

Main Results:

• The AGN model demonstrated superior bone surface extraction compared to state-of-the-art methods, achieving an Intersection over Union (IOU) of 0.92 versus 0.88.
• The system achieved near real-time performance, with an extraction speed of approximately 15 frames per second.
• Generated radiation-free 3D intraoperative bone structures suitable for intuitive orthopedic surgical navigation.

Conclusions:

• The developed annotation-guided encoder-decoder network (AGN) effectively extracts bone structures from ultrasound images.
• This method significantly reduces radiation exposure while providing accurate and efficient navigation data for orthopedic surgery.
• The AGN system offers a promising solution for enhancing safety and precision in orthopedic surgical navigation.