ConVibNet: needle detection during continuous insertion via frequency-inspired features.

Area of Science:

• Medical Imaging
• Computer Vision
• Surgical Technology

Background:

• Ultrasound-guided needle interventions are crucial but challenged by poor needle visibility.
• Existing methods struggle with artifacts, occlusions, and low contrast, limiting real-time continuous insertion support.

Purpose of the Study:

• To develop a robust real-time framework for continuous needle detection in ultrasound images.
• To overcome limitations of current approaches for accurate needle placement in interventions.

Main Methods:

• Introduced ConVibNet, an extension of VibNet, for detecting needles with reduced visibility.
• Leveraged temporal dependencies across ultrasound frames for continuous needle tip position and shaft angle estimation.
• Developed a novel intersection and difference loss to enhance temporal awareness of needle tip motion.

Main Results:

• ConVibNet demonstrated superior accuracy over baseline models (VibNet, UNet-LSTM).
• Achieved a tip error of 2.80 ± 2.42 mm and an angle error of 1.69° ± 2.00°.
• Improved tip localization accuracy by 0.75 mm while maintaining real-time inference.

Conclusions:

• ConVibNet advances real-time needle detection in ultrasound-guided interventions.
• The framework integrates temporal correlation modeling and a novel loss function for improved accuracy and robustness.
• Shows high potential for integration into autonomous needle insertion systems.