Toward Autonomous Marker Localization for Lumbar Epidural Steroid Injection Robot

Area of Science:

• Medical Imaging
• Robotics
• Computer Vision

Background:

• Accurate localization of surgical robot markers is crucial for minimally invasive procedures.
• Traditional algorithms and deep learning methods offer potential solutions for marker detection.

Purpose of the Study:

• To compare the effectiveness of Welzl's algorithm and nnU-Net for lumbar Epidural Steroid Injection (ESI) robot marker localization.
• To evaluate performance across different imaging modalities (MRI and CT).

Main Methods:

• Comparative analysis of Welzl's algorithm and 3D nnU-Net for marker localization.
• Experiments conducted on both MRI and CT datasets using identical fiducial markers.
• Benchmarking against manual and computational marking to account for human error.

Main Results:

• 3D nnU-Net demonstrated significantly superior accuracy and sensitivity on complex CT datasets compared to Welzl's algorithm.
• Welzl's algorithm outperformed nnU-Net on simpler MRI datasets.
• Localization accuracy met contingency requirements in subsequent validation experiments.

Conclusions:

• The choice of marker localization algorithm depends on the complexity of the imaging modality.
• Findings can inform improvements in clinical workflows and surgical robot registration processes.