Block diagram of the framework for needle tip localization. The input to the neural network consists of a fused image sequence derived from enhanced tip images and their corresponding B-mode images. The red arrow points to the final tip location.
Ultrasound imaging is commonly used to guide the placement of needles for tissue biopsies, catheter insertions, drainages, and anesthesia. However, visual artifacts and/or low visibility make the needle location difficult in cases of steep insertion angles (40°-80°) and depths. Hence, there is a need for a method for an accurate real-time needle detection.
Researchers at Rutgers developed a novel machine learning-based image enhancement method for improved real-time localization of needles from ultrasound data, even when the needle shaft is invisible, and the tip has a low intensity. It takes enhanced sequence of needle tip images and uses spatiotemporal information to detect the needle tip. This technology outperforms all commercially available needle enhancement platforms by ~ 30% and has a processing time of only 0.064 seconds.
This software is cohesive with current 2D and 3D cart-based and hand-held ultrasound systems for facilitated integration.
Market Applications: Ultrasound-guided procedures requiring accurate needle visualization and localization like:
- Percutaneous biopsies
- Central venous cannulation
- Regional anesthesia
- Fetal medicine
- Locates moving surgical tools at steep insertion angles and depths
- Applicable in both in-plane and out of plane imaging
- High tip localization accuracy (<1mm)
- Real-time (15fps) processing speed
- Applicable to other digital imaging techniques.
Intellectual Property & Development Status: Patent pending. Available for licensing and/or collaboration.
Publication: Hacihaliloglu I., Time-aware deep neural networks for needle tip localization in 2D ultrasound. Int J CARS (2021)