Enhanced needle localization in ultrasound using beam steering and learning-based segmentation.

Area of Science:

• Medical Imaging
• Machine Learning
• Surgical Navigation

Background:

• Needle segmentation in ultrasound imaging is crucial for accurate guidance during medical procedures.
• Current methods face challenges in precise needle localization and orientation detection.
• Improving visualization and targeting accuracy is essential for patient safety and procedural success.

Purpose of the Study:

• To develop and validate a machine learning-based method for segmenting needles in 2D beam-steered ultrasound images.
• To enhance the accuracy and success rate of needle localization compared to existing techniques.
• To assess the method's robustness with and without prior knowledge of needle orientation.

Main Methods:

• Utilized a statistical boosting approach for pixel-wise needle segmentation.
• Employed the Radon transform to determine needle position and orientation from segmented images.
• Validated the method on ex vivo specimens and clinical nerve block procedures.

Main Results:

• Achieved 86.2% localization success with a 0.48mm mean targeting error on ex vivo data (a priori orientation known).
• Demonstrated robustness to the lack of a priori needle orientation knowledge.
• Reached 99.8% localization success with a 0.19mm mean targeting error on clinical data (orientation aligned with beam steering).

Conclusions:

• The proposed learning-based segmentation method significantly improves needle localization accuracy and success rates.
• Enhanced targeting accuracy and visualization can benefit ultrasound-guided needle procedures.
• This technique holds potential for increasing precision and safety in interventions requiring needle guidance.