Needle Detection and Localization in 3D Ultrasound Data Volumes

Background: 

Ultrasound-guided interventions are broadly applied in medical procedures, e.g. for regional anesthesia or ablation. . However, for a typical 2D ultrasound (US) system, bi-manual coordination of the needle and US transducer for maintaining the perfect alignment is challenging and an inadequate view of the tool leads to an erroneous placement. For this reason, medical specialists need considerable practicing and training to increase the success of interventions. Using a 3D transducer and an automated needle tracking device, the instrument is conveniently placed in the 3D field of view and the processing unit automatically localizes and visualizes the entire needle to the medical specialist. Recent developments in image-based needle detection are shown to be appealing as the important components and devices (e.g. needle, signal generation and transducer) of the system remain unaltered.

Description: 

Image-based detection of the needle is investigated using supervised classification of voxels with hand-crafted features, convolutional neural networks and unsupervised localization of the needle planes. However, further improvements are still possible. These methods require extensive computational power and are not yet suitable for real-time applications. In order to increase the robustness of the needle detection system in challenging datasets and limit the computational complexity, novel processing techniques need to be investigated. Possible solutions include but not limited to:

  • Novel pre-processing methods for improving ultrasound data quality and removing artifacts
  • Investigating novel machine learning frameworks to robustly classify needle voxels in 3D
  • Novel alternative configuration of US systems for needle detection in 3D

For this project, experience with either C++ or Python programming languages is a must.

Location: 
This project will be executed internally at the TU/e. There will be progress meetings with partners in Catharina hospital and Philips research.
Keywords: 
Machine learning, Image analysis, Computer-Aided Diagnosis (CAD), Ultrasound, Computer-Assisted Interventions
Contact person: 
Partners: