Ultrasound-based liver tracking utilizing a hybrid template/optical flow approach

Area of Science:

• Medical imaging
• Surgical technology
• Biomedical engineering

Background:

• Minimally invasive surgery requires precise tracking of internal targets.
• Non-invasive treatments like stereotactic radiation therapy and high-intensity focused ultrasound depend on accurate target localization.
• Current methods for tracking moving targets in ultrasound images need improvement for optimal treatment delivery.

Purpose of the Study:

• To develop a robust, accurate, and fast method for real-time target tracking using ultrasound images.
• To enhance the precision of target localization for image-guided medical interventions.
• To address the need for improved tracking in minimally invasive procedures.

Main Methods:

• A novel algorithm combining template matching, dense optical flow, and image intensity information was developed.
• A weighting map was generated from each component, normalized, weighted, and combined to predict target position.
• The method was evaluated on the Challenge for Liver Ultrasound Tracking 2015 dataset.

Main Results:

• The algorithm achieved an accuracy of 1.91 mm on training data and 1.85 mm on test data.
• Tracking frequencies ranged from 8 to 36 frames per second (mean 22 fps), depending on the region of interest size.
• The proposed method demonstrated a mean accuracy improvement of approximately 0.3 mm compared to existing literature.

Conclusions:

• An accurate and robust method for tracking points of interest in 2D ultrasound data was successfully developed.
• The combination of multi-template matching, dense optical flow, and image intensity information provides superior tracking performance.
• This technique holds significant potential for improving image-guided surgeries and non-invasive treatments.