Automated Multiple Needle Detection For Brachytherapy


Workflow to detect multiple needles in three-dimensional ultrasound images.

Key Benefits

  • Does not require specialized ultrasound pulses or hardware.
  • Detects multiple needles with low error.

Market Summary

Many minimally invasive procedures require inserting a needle to access a target site inside the patient body, avoiding the requirement of making large incisions. Brachytherapy is a cancer treatment procedure that involves placing radioactive material inside the patient’s body. Accurate placement of the brachytherapy needle is vital to the intervention, as incorrect needle insertion can lead to a failure of the procedure and increase the risk of complications. Ultrasound (US) is broadly adopted to visualize and guide the interventions because it is safe, low cost and real-time. However, needle detection in US images remains a challenging problem due to the low signal-to-noise ratio and image artifacts of US imaging.

Technical Summary

Emory University researchers have developed a sparse dictionary machine learning algorithm called "order-graph regularized dictionary learning (ORDL)." This algorithm first pre-processes 3D ultrasound images with and without needles. These images are used to train the ORDL machine learning algorithm. Once the dictionary is trained, needles can be reconstructed on the target image and then detected with a random sample consensus (RANSAC) algorithm. This technique may facilitate the clinical workflow by providing rapid and accurate needle detection required for treatment planning in US-guided brachytherapy.

Developmental Stage

Experiments have been conducted on phantom data set and can correctly detect 95% of needles with a tip error of 1.01 mm.

Patent Information

Tech ID: 19248
Published: 11/2/2020