RSV: Robotic Sonography for Thyroid Volumetry

John Zielke, Christine Eilers, Benjamin Busam, Wolfgang Weber, Nassir Navab, Thomas Wendler

Introduction

In the treatment of hyperthyroidism through radioiodine therapy (RIT), the precision of the radiation dose is paramount. This dosage is fundamentally dependent on the volume of the patient's thyroid gland. Currently, clinical practice relies on 2D ultrasound scans and a simplified "ellipsoid formula" to estimate this volume. However, this manual approach is highly user-dependent and often fails to account for the complex, asymmetrical geometry of the thyroid, leading to significant dosing errors. Under-treatment fails to resolve the disease, while over-treatment can cause unnecessary damage to surrounding healthy tissue. To address these inconsistencies, researchers have developed Robotic Sonography for Thyroid Volumetry (RSV), an automated system designed to standardize and improve the accuracy of thyroid volume measurements.

Methodology

The RSV system integrates a 6-degree-of-freedom (DOF) KUKA LBR iiwa robotic arm equipped with a linear ultrasound probe. The workflow is divided into two distinct phases: robotic acquisition and volumetric estimation.

1. Robotic Acquisition

The process begins with a human operator placing the probe at an initial starting point on the patient's neck. Once initiated, the robot utilizes impedance control to maintain a constant, gentle contact pressure against the skin. Unlike static robotic systems, RSV employs an "online" feedback loop using the QuickNAT deep learning network to segment the thyroid in real-time. This allows the robot to adjust its trajectory dynamically:

  • Shadow Prevention: If the system detects acoustic shadows (indicating poor contact or interference), it automatically rotates the probe to restore image quality.
  • Lobe Centering: If the thyroid lobe begins to drift toward the edge of the ultrasound frame, the robot shifts laterally to keep the organ centered, ensuring a complete data sweep.

2. Volumetry Estimation

Following the scan, the collected 2D slices are compounded into a 3D volume. A deep neural network segments the thyroid across all frames, and the final volume is calculated by counting the voxels (3D pixels) within the identified boundaries. This method moves beyond the geometric approximations of the ellipsoid formula, providing a more direct measurement of the organ's actual shape.

Experimental Setup and Datasets

The system was validated using a CIRS Model 074 thyroid ultrasound phantom, which replicates the acoustic properties of human tissue. To establish a definitive ground truth, the researchers performed a high-resolution CT scan of the phantom, manually segmenting it to determine a true volume of 30.01 ml. The QuickNAT segmentation network was specifically trained on phantom-derived images to ensure high-precision tracking during the robotic trials.

Key Results and Insights

The RSV system was tested against 12 human volunteers, including 5 experts and 7 non-experts, who performed traditional manual 2D scans for comparison.

  • Accuracy Gains: For expert clinicians, the mean volume measurement error dropped from 20.85% using manual methods to 8.23% with the robotic system.
  • Democratizing Expertise: The most significant impact was observed among non-experts. The robotic system reduced their mean error by 85% compared to manual 2D methods, bringing their performance to a level that rivals or exceeds experienced clinicians.
  • Repeatability: The system demonstrated high consistency, achieving a standard deviation of only 1.00 ml when utilizing both shadow prevention and centering features.
  • The Value of Image Guidance: An ablation study confirmed that a simple linear robotic scan (without real-time adjustments) performed poorly. The ability to "close the loop" by using live ultrasound data to guide motion is essential for navigating the curved anatomy of the neck.

Limitations and Future Directions

While the RSV system represents a significant step toward automated diagnostics, several challenges remain before clinical implementation. The current neural network is optimized for a specific ultrasound phantom; real human tissue presents significantly higher variability in texture and anatomy. Furthermore, human patients introduce dynamic variables such as breathing, swallowing, and complex skeletal structures like the clavicle. Future iterations of the system will need to incorporate multi-directional force analysis and more sophisticated path-planning algorithms to accommodate patients with enlarged thyroids (goiters) and other pathological variations.

Summary updated

Abstract

In nuclear medicine, radioiodine therapy is prescribed to treat diseases like hyperthyroidism. The calculation of the prescribed dose depends, amongst other factors, on the thyroid volume. This is currently estimated using conventional 2D ultrasound imaging. However, this modality is inherently user-dependant, resulting in high variability in volume estimations. To increase reproducibility and consistency, we uniquely combine a neural network-based segmentation with an automatic robotic ultrasound scanning for thyroid volumetry. The robotic acquisition is achieved by using a 6 DoF robotic arm with an attached ultrasound probe. Its movement is based on an online segmentation of each thyroid lobe and the appearance of the US image. During post-processing, the US images are segmented to obtain a volume estimation. In an ablation study, we demonstrated the superiority of the motion guidance algorithms for the robot arm movement compared to a naive linear motion executed by the robot in terms of volumetric accuracy. In a user study on a phantom, we compared conventional 2D ultrasound measurements with our robotic system. The mean volume measurement error of ultrasound expert users could be significantly decreased from <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$20.85\pm 16.10\%$</tex-math></inline-formula> to only <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$8.23\pm 3.10\%$</tex-math></inline-formula> compared to the ground truth. This tendency was observed even more in non-expert users where the mean error improvement with the robotic system was measured to be as high as 85% which clearly shows the advantages of the robotic support.
BibTeX
@article{zielke2021rsv,
  author = {John Zielke and Christine Eilers and Benjamin Busam and Wolfgang Weber and Nassir Navab and Thomas Wendler},
  title = {{RSV: Robotic Sonography for Thyroid Volumetry}},
  journal = {IEEE Robotics Autom. Lett.},
  year = {2021},
  doi = {10.1109/lra.2022.3146542},
}
Figure