A New Registration/Visualization Paradigm for CT-Fluoroscopy Guided RF Liver Ablation

Ruxandra Micu, Tobias F. Jakobs, Martin Urschler, Nassir NavabMICCAI 2006 · DOI

Introduction

Radiofrequency Ablation (RFA) has emerged as a vital minimally invasive treatment for liver tumors, utilizing thermal energy delivered via a needle electrode to destroy malignant tissue. The success of this procedure hinges on the precise placement of the electrode, which is typically guided by real-time CT-fluoroscopy (CT-fluoro). However, CT-fluoro presents a significant challenge: to maintain patient safety regarding radiation, the images are often noisy, low-contrast, and limited to a single 2D slice. This makes it difficult for interventionalists to accurately map the real-time view to high-resolution pre-operative 3D CT scans.

The Challenge of Liver Registration

Aligning a grainy 2D slice with a detailed 3D volume is complicated by several factors. First, the liver is a highly mobile and deformable organ; breathing causes it to shift significantly between the pre-operative breath-hold scan and the active procedure. Second, the liver's relatively uniform texture provides few natural landmarks for automated alignment. Finally, the metallic ablation needle itself creates significant streak artifacts in CT images, which can mislead standard registration algorithms.

A Piecewise Registration Strategy

Rather than pursuing the computationally intensive and often unstable goal of a single, perfect non-rigid registration, this research proposes a "piecewise" approach. Recognizing that the ventral (front) and dorsal (back) portions of the liver move differently during respiration, the researchers split the liver into two sections at the ligamentum falciforme.

Independent rigid (6 degrees of freedom) or affine (12 degrees of freedom) registrations are then applied to each half. To handle needle artifacts, the system employs a robust "bell-shaped" intensity-based similarity measure designed to ignore high-intensity outliers. This allows the algorithm to focus on the underlying liver tissue rather than the surgical instruments.

Validation and Visualization

To test the efficacy of this approach, the authors utilized three distinct datasets, including real patient data and simulated data designed to provide a "ground truth" for error measurement. The simulation involved applying mathematical models of breathing deformation and low-dose radiation noise to high-resolution CT scans.

Comparison of real CT-fluoroscopy slices (a, c) and simulated data (b, d) used to establish ground truth for registration validation.

Beyond registration accuracy, the paper introduces a new visualization paradigm. Instead of displaying a single static slice, the system calculates two optimal planes (one for each liver section) and generates an animation that rotates through the sub-volume between them. This provides the physician with a better sense of the 3D neighborhood surrounding the tumor and the needle.

Key Results and Clinical Impact

The piecewise registration approach demonstrated significant improvements in accuracy. In simulated tests, using two registration planes nearly halved the Root Mean Square (RMS) error compared to a single-plane approach, dropping from approximately 10.2mm to 5.5mm.

Clinical feedback from experienced interventional radiologists was highly positive. The animated sub-volume visualization was noted as particularly valuable for spatial orientation, potentially reducing the need for doctors to leave the sterile field to consult 3D workstations. While the current system requires some manual input to define the liver sections, it represents a major step toward a more reliable and context-aware guidance system for liver RFA.

Summary updated

BibTeX
@inproceedings{micu2006new,
  author = {Ruxandra Micu and Tobias F. Jakobs and Martin Urschler and Nassir Navab},
  title = {{A New Registration/Visualization Paradigm for CT-Fluoroscopy Guided RF Liver Ablation}},
  booktitle = {International Conference on Medical Image Computing and Computer-Assisted Intervention},
  year = {2006},
  doi = {10.1007/11866565\_108},
}
Figure