Poster R15 - Image Guidance and Safe Insertion Region Localization for Lumbar Puncture using Ultrasound

Introduction: Lumbar Puncture (LP), or Spinal Tap, is a medical procedure used to access cerebrospinal fluid (CSF) for diagnostic and therapeutic purposes. Approximately 363,000 LPs are performed each year among ED-treated patients in the United States, and every year approximately 400,000 diagnostic lumbar punctures are performed by neurologists in the United States. Traditionally, this procedure relies on manual needle insertion guided by ultrasound and palpation of anatomical landmarks. The insertion is into the interspinal spacing between two spinous processes of the lumbar vertebrae. This interspinal space typically ranges from about 1 to 3 centimeters (cm). This can make the procedure complex and increase the learning curve. Misalignment of the needle can cause traumatic lumbar puncture or leakage of CSF, which has side-effects include headaches, cranial neuropathies, prolonged backache, nerve root injury, and meningitis. We believe that effective visualization can simplify the learning process and enhance the workflow in lumbar puncture procedure. Our work proposes the integration of 2D ultrasound imaging with 3D MRI model, aiming to create a real-time visualization tool for the lumbar spine and entry locations which might help in bringing ease to the puncture procedure.

Materials and

Methods: The process was divided into two phases: the Registration Phase and the Dynamic Scanning Phase. In the registration phase, the Ultrasound Lumbar Puncture Simulator IIA (Kyoto Kagaku) was scanned using a GE Logiq E9 Ultrasound Machine with a Linear Array Probe, capturing 20-25 images over approximately one minute. An intensity based image processing model was developed for identifying the edges of the spinous processes in the ultrasound images. A VICON motion capture system tracked the probe poses, and using corresponding image points, a sparse ultrasound point cloud was constructed. This point cloud was initially aligned with a ground truth MRI point cloud using the spinous process tips of both modalities as 4-4 correspondences. This alignment was refined using the Iterative Closest Point Algorithm and a transformation matrix between MRI ground truth and ultrasound point cloud was obtained. In the dynamic scanning phase, the established transformation train is used to overlay MRI-defined safe regions onto real-time ultrasound images. To overcome errors in the registration due to inaccurate edge detection or motion of the body, a local optimization technique was introduced into the dynamic scanning pipeline. For every real-time ultrasound image captured, the probe pose is localized on the MRI ground truth, and an ultrasound equivalent slice of MRI is taken. In local optimization, the distances between edges of ultrasound image and MRI slices are minimized while updating the registration between them.

Results, Conclusions, and Discussions: The registration, transformation, and local optimization is used for real-time overlay of predefined "safe regions" from the MRI model onto ultrasound images. The transformation between MRI and Ultrasound point clouds is continuously updated by local optimization and Intersection of Union (IoU) percentages are used to track alignment quality. Testing on 44 dynamically scanned ultrasound images showed an improvement from an average IoU of 72.15% before optimization to 85.35% after optimization. This system establishes a framework for merging ultrasound images with MRI ground truth to provide a visual guide to lumbar puncture procedures. Future developments will focus on defining needle entry and target points directly on MRI scans to facilitate precise, real-world insertion guidance. We can also explore the use of a generalized 3D spine model for broader applicability, alternative tracking technologies for clinical environments, and adjustments to accommodate multi-directional patient movements. These advancements can potentially reduce procedural learning curves and improve outcomes significantly.

Acknowledgements (Optional):