Automatic and accurate needle detection in 2D ultrasound during robot-assisted needle insertion process

Area of Science:

• Medical Robotics
• Image-guided Surgery
• Artificial Intelligence in Medicine

Background:

• Robot-assisted needle insertion using 2D ultrasound (US) improves puncture accuracy and success rates.
• Accurate needle tracking is crucial for monitoring procedures, preventing deviations, and minimizing tissue damage.
• Existing methods require robust solutions for real-time needle detection in US images.

Purpose of the Study:

• To develop an automated framework for precise needle detection in 2D ultrasound images during robot-assisted insertion.
• To enhance the safety and efficiency of US-guided minimally invasive procedures through improved needle tracking.

Main Methods:

• Proposed a novel convolutional neural network (CNN) with a two-channel encoder and single-channel decoder for needle segmentation.
• Utilized needle motion information from adjacent US frames for enhanced segmentation.
• Developed an automatic detection framework employing a region of interest (ROI) prediction for continuous, faster localization.

Main Results:

• The needle segmentation network achieved high accuracy (99.7%), precision (86.2%), recall (89.1%), and F1-score (0.87).
• The detection framework demonstrated precise needle localization with a mean tip error of 0.45 ± 0.33 mm and orientation error of 0.42° ± 0.34°.
• Achieved a rapid processing time of 50 ms per image, enabling real-time application.

Conclusions:

• The developed framework enables robust, accurate, and real-time needle localization in robot-assisted procedures.
• Shows significant promise for enhancing needle tracking and safety in challenging US-guided minimally invasive interventions.
• Facilitates safer and more effective robotic-assisted automatic puncture procedures.