Multi-Needle Detection in 3D Ultrasound Images Using Unsupervised Order-Graph Regularized Sparse Dictionary Learning.

Area of Science:

• Medical Imaging
• Computational Imaging
• Biomedical Engineering

Background:

• Accurate multi-needle detection in 3D ultrasound (US) is crucial for US-guided brachytherapy treatment planning.
• Current methods often focus on single-needle detection using limited data, neglecting vast US image datasets without needles.

Purpose of the Study:

• To develop an automated workflow for multi-needle detection in 3D US, utilizing images without needles as auxiliary data.
• To improve the accuracy and efficiency of needle localization for image-guided procedures.

Main Methods:

• Developed an enhanced sparse dictionary learning method (order-graph regularized dictionary learning) integrating spatial continuity of 3D US.
• Trained position-specific dictionaries on 3D patches from auxiliary images (without needles).
• Reconstructed target images, clustered residual pixels to find centers, constructed ROIs, and applied random sample consensus for needle detection and tip localization.

Main Results:

• The proposed workflow achieved 95% needle detection accuracy on a prostate dataset.
• A needle tip localization error of 1.01 mm was reported on the prostate dataset.
• Experiments on phantom and patient datasets demonstrated the workflow's effectiveness.

Conclusions:

• The novel workflow enables accurate multi-needle detection in 3D US, significantly aiding treatment planning for US-guided brachytherapy.
• This technique facilitates clinical workflow by providing precise needle localization for procedures like HDR prostate brachytherapy.